The current issue and full text archive of this journal is available at http://www.emerald-library.com BIJ 7,1 8 Methods and techniques to help quality function deployment (QFD) Vivianne Bouchereau and Hefin Rowlands University of Wales College, Newport, South Wales, UK Keywords Quality function deployment, House of quality, Fuzzy logic, Neural networks, Taguchi methods Abstract Quality function deployment (QFD) is a management tool that provides a visual connective process to help teams focus on the needs of the customers throughout the total development cycle of a product or process. It provides the means for translating customer needs into appropriate technical requirements for each stage of a product/process-development life-cycle. It helps to develop more customer-oriented, higher-quality products. While the structure provided by QFD can be significantly beneficial, it is not a simple tool to use. This article outlines how techniques such as fuzzy logic, artificial neural networks, and the Taguchi method can be combined with QFD to resolve some of its drawbacks, and proposes a synergy between QFD and the three methods and techniques reviewed. Benchmarking: An International Journal, Vol. 7 No. 1, 2000, pp. 8-19. # MCB University Press, 1463-5771 Introduction To succeed in developing thriving new products or improve on existing ones is not easy. Studies indicate that as much as somewhere between 35 per cent and 44 per cent of all products launched are considered failures (Urban, 1980). It is one thing to actually discover and measure the customers' needs and wants but, to achieve results, these findings need to be implemented, i.e. translated into company language. Many companies depend on their warranty programmes, customer complaints, and inputs from their sales staff to keep them in touch with their customers (Akao, 1990). The result is a focus on what is wrong with the existing product or service, with little or no attention on what is right or what the customer really wants. The success of a product or service largely depends on how they meet the customers' needs and expectations. Consequently, more effort is involved in getting the information necessary for determining what the customer truly wants. This tends to increase the initial planning time in the project definition phase of the development cycle, but it reduces the overall cycle time in bringing a product to market. One process-oriented design method constructed to carry out the translation process and make sure that the findings are implemented is quality function deployment (QFD). QFD is a visual connective process that helps teams focus on the needs of the customers throughout the total development cycle. It provides the means for translating customer needs into appropriate technical requirements for each stage of a product/process development life-cycle. It is well documented that the use of QFD can reduce the development time by 50 per cent, and start-up and engineering costs by 30 per cent (Clausing and Pugh, 1991). While the structure provided by QFD can be significantly beneficial, it is not a simple tool to use. It is a complex and very Quality function time-consuming process to develop the QFD charts. Among its drawbacks are deployment the complexities of its charts, the vagueness in the data collected and the analysis is performed on a rather subjective basis. This article addresses these issues and gives a review of potential techniques and methods to overcome these problems. Artificial intelligence techniques such as fuzzy logic and 9 artificial neural networks, together with management and statistical tools such as the Taguchi method are proposed to resolve some of QFD's drawbacks. The article gives a brief introduction to QFD, together with its advantages and disadvantages. Then fuzzy logic is reviewed and an introduction to how it can be incorporated within QFD is highlighted. Artificial neural networks are then considered together with what role it can play in helping QFD. As a method to help benchmarking in the house of quality, the Taguchi method is introduced. Conclusions are then made regarding how to integrate all the techniques and methods together to produce an intelligent systems approach to QFD. QFD QFD was originated in the late 1960s to early 1970s, in Japan, by Professor Yoji Akao (Akao, 1972). It is an integrated set of tools for recording user requirements, engineering characteristics that satisfy these user requirements, and any trade-offs that might be necessary between the engineering characteristics. Many definitions of QFD have been proposed which reflects its many facets. However, QFD is primarily a people system. Nothing happens without people. Its point of departure is the ``voice of the customers'' (VOC). It also brings together multifunctional teams to work together towards satisfying the customer. QFD also helps to build a partnership between customers and suppliers. Companies are sometimes too internally focused and therefore develop goods or services with a vague understanding of the customers' requirements, or they are too externally focused, trying to constantly please the customer at the expense of their own business survival. QFD can help companies make the key trade-offs between what the customer wants and what the company can afford to build. By concentrating efforts on what will satisfy the customers and the company most, less time will be spent on redesign and modification of the product/process. It helps companies to move from an inspection-based approach to designing quality into products and as such plays a key role in any total quality management (TQM) or continuous improvement programme or implementation. QFD does nothing that people did not do before, but it replaces inconsistent, intuitive decision-making processes with a structured approach. The QFD process The starting point of any QFD project is the customer requirements, often referred to as the non-measurable such as ``how it looks, how it feels, durability, BIJ 7,1 10 etc.''. These requirements are then converted into technical specifications like ``oven temperature, mould diameter, etc.''. This stage is referred to as the engineering characteristics or measurables. The QFD process involves four phases: (1) Product planning: house of quality. (2) Product design: parts deployment. (3) Process planning. (4) Process control (quality control charts). A chart (matrix) represents each phase of the QFD process. The complete QFD process requires at least four houses to be built that extend throughout the entire system's development life-cycle (Figure 1), with each house representing a QFD phase. In the first phase, the most important engineering characteristics, that satisfy most of the customers' demands defined by the scoring at the bottom of the house, go on to form the input to the subsequent stage in the QFD process. The house of quality The first chart is normally known as the ``house of quality'', owing to its shape (Figure 2a). Figure 2b shows an example of the house of quality for a paper-roll manufacturing process. The QFD charts help the team to set targets on issues, which are most important to the customer and how these can be achieved technically. The ranking of the competitors' products can also be performed by technical and customer benchmarking. The QFD chart is a multifunctional tool that can be used throughout the organization. For engineers, it is a way to summarise basic data in a usable form. For marketing, it represents the customer's voice and general managers use it to discover new opportunities (Clausing and Pugh, 1991). Figure 1. The four phases of QFD Quality function deployment 11 Figure 2. (a) House of quality, (b) house of quality for a paper-roll manufacturing process BIJ 7,1 12 Benefits Drawbacks Customer-oriented Brings together large amounts of verbal data Ambiguity in the VOC Need to input and analyse large amounts of subjective data QFD development records are rarely kept Manual input of customer survey into the house of quality (HOQ) is time-consuming and difficult QFD analyses often stop after the first HOQ, so links between the four QFD phases are broken The HOQ can become very large and complex Setting target values in the HOQ method is imprecise Strength of relationship is ill-defined Brings together multi-functional teams Reduces development time by 50 per cent and reduces start-up and engineering cost by 30 per cent Helps design quality into the products at the design stage Organizes data in a logical way Table I. Benefits and drawbacks of QFD QFD is used not only for products, but for processes and services as well Strengthens good relationship between customer and company Improves customer satisfaction QFD is a qualitative method Benefits and drawbacks of QFD Companies which attempt to implement QFD have reported a variety of benefits and problems with the method. Table I summarises some benefits and drawbacks of QFD. These drawbacks have prompted the need for new approaches to the application of the QFD method. Combining QFD with other techniques, such as fuzzy logic, artificial neural networks (ANN) and the Taguchi method, will help address these issues and forms the basis of future research in this field. Fuzzy logic Various inputs, in the form of judgements and evaluations, are needed in the QFD charts. Normally, the marketing department through questionnaires, interviews and focus groups collects these inputs. This gives rise to uncertainties when trying to quantify the information. In order to reduce the uncertainty in the data collected, fuzzy logic can be used. Fuzzy logic can model vagueness in data and/or relationship in a formal way. This technique is able to manipulate fuzzy qualitative data in terms of linguistic variables. Professor L.A. Zadeh introduced fuzzy logic and fuzzy sets in 1965 (Zadeh, 1965). Fuzzy logic uses human linguistic (words and sentences) understanding to express the knowledge of a system. This knowledge consists of facts, concepts, theories, procedures and relationships and is expressed in the form of IF-THEN rules. Linguistic variables are characterised by ambiguity and multiplicity of meaning. Specifying good linguistic variables depends on the knowledge of the expert. For example, ``age'' is a linguistic variable if its values are ``young'', ``not so young'', ``old'' and ``very old''. In fuzzy logic theory, a linguistic variable can be a member of more than one group. For instance, someone who is 27-years- Quality function old belongs to both the ``young'' and ``not so young'' group to a different degree deployment as can be seen in Figure 3. Fuzzy logic exhibits some useful features for exploitation in QFD. These include: . . it uses human linguistic understanding to express the knowledge of the system; 13 it allows decision making with estimated values under incomplete or uncertain information; . it is suitable for uncertain or approximate reasoning; . interpretation of its rules is simple and easy to understand; and . it deals with multi-input, multi-output systems. Membership Function Integrating fuzzy logic with QFD The VOC in the QFD process is usually expressed in the customer's own words, which can be interpreted in a linguistic form. The VOC usually comes in qualitative forms; however, their performance measures and other associated data should as far as possible be expressed quantitatively to facilitate further downstream analysis. Essentially, the VOC contains ambiguity and different meanings. The adjectives in particular are not specific. ``The product must be able to last a long time'' and ``part size must be small'' are some examples. Various inputs, in the form of judgements and evaluations, are needed during the QFD analysis. A particularly difficult task is the subjective decisions that have to be made when correlating the customer's demands to the engineering characteristics. Fuzzy logic has the ability to deal with subjective decisions and is particularly suited as a quantitative method to evaluate these subjective decision-making processes. In the fuzzy logic-based QFD approach, symbols which represent the customer demands and engineering characteristics (strong, medium, weak), are used to fill the relationship matrix and build the house of quality (Room 4 in Not so Young Young 1.0 Old Very Old 0.5 age (years) 0 0 20 27 40 60 80 Figure 3. Fuzzy logic representation of age BIJ 7,1 14 Figure 2a) (Khoo and Ho, 1996). The symbol descriptions are normally in the form of linguistic variables. These linguistic variables can be translated into fuzzy numbers, as shown in Table II. Instead of using exact values, the range of values, which are more natural, can be used to represent the vagueness in these three relationships, that is ``strong'', ``medium'' and ``weak''. Two propositions for using fuzzy sets in QFD are developed in Masud and Dean's (1993) study. In both situations, all the input variables are regarded as linguistic variables with values shown as linguistic expressions. Each of these expressions is then converted to a fuzzy number. The difference between the two approaches is concerned with how the QFD calculations are performed with these fuzzy numbers. What the methods show is that fuzzy logic can prove useful in interpreting subjective data into a more quantitative format that can be used in the decision-making process in QFD. Another area of the house of quality, the customer evaluation of the in-house and competitors' product (Room 6), has also benefited from fuzzy logic and fuzzy set theory (Wasserman et al., 1993). Their work gives details on how to construct an overall customer satisfaction index to determine the best product among the competitors based on the use of the technique for order preference by similarity to ideal solution (TOPSIS). They suggest that quantifying the customer satisfaction of the competitive product is not easily done on a linear scale, as the information contains linguistic information. To resolve this difficulty, they use the conversion scales proposed by Cheng and Hwang (1992) to convert linguistic terms into their fuzzy equivalents. Hence, the fuzzy set framework is adopted to transform linguistic data to crisp score as opposed to directly using the linear scale of the customer response and multi-attribute decision making (MADM) which is then used to calculate overall customer preferences. ANN ANN can be considered as simplified mathematical models of the human brain which function as computing networks (Hammerstrom, 1993). They consist of simple processing elements called ``neurons'', that exchange signals along ``weighted'' connections (Figure 4). ANN makes use of the way that the human brain learns and functions and represents this information in mathematical algorithms incorporated in computers. They possess the ability to learn from Linguistic variables Table II. Definition of linguistic variables Strong relationship Moderate relationship Weak relationship Fuzzy number [4.0, 10.0] [2.0, 8.0] [0.0, 6.0] examples and thus have the ability to manage systems from their observed Quality function behaviour rather than from a theoretical understanding. This ability to learn deployment from experience is very useful in the real world. ANN exhibit some valuable features that can be useful for merging it with QFD. These include: . the ability to deal with a large amount of input data; 15 . the ability to deal with imprecise data and ill-defined activities ± they can tolerate faults; . they are adaptive, possessing the ability to learn from examples; . they can reduce development time by learning underlying relationships; and . they are non-linear, that is they can capture complex interactions among the input variables in a system. Integrating ANN and QFD Another major drawback of QFD is the need to deal with large amounts of data gathered from the customers, competitors, engineers, etc., and it calculates values on a rather subjective basis. The ability of the neural network to generalise functional relationships among example data is of great importance for design. This property is important wherever these functional relationships are assumed, but not known. Owing to it being able to mimic so many human behaviours, ANN is well-suited for integration with QFD. Zhang et al. (1996) have proposed a machine-learning approach to QFD, in which a neural network automatically evaluates the data by learning from examples. Customer demands, engineering characteristics and engineering solutions are interconnected as shown in Figure 4 and represented in a neural network format. Engineering solutions are considered as the input, and customer satisfaction rating as the output. Each neuron represents a node (e.g. engineering solution is a node) and each link between neurons represents a relationship (e.g. there are relationships between engineering characteristics and customer demands). Figure 4. Artificial neural networks: design theory and their interrelationship BIJ 7,1 16 The suggestion is to incorporate the engineering solutions of the product (the in-house and the competitor's product), within the neural network to find weighting that represents the customer's satisfaction. This result will derive values in Room 6 (Figure 2a) of the house, instead of customers subjectively ranking the competitors' and in-house products. The Taguchi method The Taguchi method is a combination of an engineering approach and a statistical method to achieve improvements in product/process's cost and quality, accomplished through design optimisation. Dr Genichi Taguchi developed the Taguchi method in the early 1960s in Japan (Taguchi, 1986). His method is based on the design of experiments (DOE) to provide near-optimal quality performance. The goal is to identify parameters that can be controlled (control factors) and to reduce the sensitivity of engineering designs to uncontrollable factors (noise). This is achieved by using small-scale experiments in the laboratory to find reliable designs for large-scale production. Instead of defining quality as a positive attribute of a product, it is defined as a financial loss or cost to society cause by undesired variance in the product. Dr Taguchi has been particularly recognised for three major contributions to the field of quality (Taguchi, 1993): (1) the quality loss function; (2) orthogonal arrays; and (3) robustness. Some of the benefits of the Taguchi method can prove useful for exploitation in QFD. These include: . . . . the modelling of interactions between characteristics, useful for the roof area of the house of quality; the optimization of target values using the loss function, useful for setting target values after customer and technical benchmarking in QFD; determining the nature of relationships between demands and optimise the conflicts; and helping to design robust products that are insensitive to variations in environmental conditions. The Taguchi method helps QFD Most QFD applications stop after one matrix, the house of quality. Of the few applications that reach deployment into manufacturing, determination of the best manufacturing conditions is not a precise process. Taguchi's philosophy of robust design is particularly useful for establishing the best operating Quality function conditions for manufacturing and can thus be integrated in the third QFD deployment phase, the process planning phase (Figure 1). Terninko (1992) has proposed that the concept of Taguchi's quality loss function offers an improved way to accomplish technical benchmarking at the bottom of the house of quality. Before benchmarking, the team is really 17 dealing with customer perceptions and not actual performance. Technical benchmarking is necessary to rationally select target values for performance measures. Data collected for technical benchmarking should be gathered in a real environment. QFD attempts to do just that by going to the Gemba, that is the total environment where the customer lives and works. Different customer environments can be used to find the average performance and the variation of a product/process. As part of the house of quality, the customers and engineers evaluate both their product/process against that of the competitors to help determine the approximate target value. Identifying the target value is not an easy task. Targets are sometimes the designer's best guess. The quality loss function curve, which is used to measure how satisfied the customers are in financial terms, is centred on the target and can help determine the exact target values in the house of quality. Existing explanations of QFD assume that the customer requirements (the WHATs) are constant, i.e. either as unchanging over time or as the same for all customers at a given point. It does not address those situations where they are dynamic. Re Velle (1991) suggests that customer requirement are dynamic and cannot be controlled by a supplier. Using the Taguchi inner-outer array table, a method to identify the most robust engineering characteristics (the HOWs) to satisfy the range of customer importance rating is presented. The outer array is used to represent the customer demands (WHATs) with the corresponding orthogonal array. In this way the customer demands are treated as noise. The engineering characteristics (HOWs) are represented as an inner array. The customer then tests all the different combinations of the product and a customer agreement index is placed in the resultant matrix. The signalto-noise (S/N) ratio (bigger is better, nominal is best or smaller is better) is then calculated. The predicted value of the S/N ratio is then used to identify the most robust parameters at optimal factor level. As a result, a robust requirement matrix is created, which is insensitive to changes in the needs of the customer. Conclusion After a period of increasing professionalism in almost all areas of research and technology, the new era is to embrace the interaction of different approaches, to form multi-functional disciplines. It is time to take advantage of other methods and to incorporate them within the QFD process, to realise its full potential. QFD is a planning tool and organizes data in a logical and systematic way, but it is rather a qualitative method. The union of QFD with quantitative methods Figure 5. A proposed architecture that will integrate QFD, fuzzy logic, the Taguchi method and artificial neural networks Important characteristics III Process Planning Important characteristics TAGUCHI METHODS (Loss Function conceptreduction in variation) TAGUCHI METHODS (Loss Function conceptreduction in variation) TAGUCHI METHODS (Loss Function conceptreduction in variation) Design of Experiment (DOE) Design of Experiment (DOE) Design of Experiment (DOE) (OPTIMISE) (OPTIMISE) (OPTIMISE) Production Requirements Key Process Operation Key Process Operation Parts characteristics II Product Design Artificial Neural Network Important characteristics Parts Characteristics FUZZY LOGIC/ I Product Planning FUZZY LOGIC/ Customer Demands Engineering Characteristics Engineering characteristics 18 will yield even greater benefits from its application. This article has given an overview of QFD together with its advantages and limitations. Some of its limitations have prompted research into techniques that can help to improve the QFD process. The application of the theory of fuzzy logic provides a more quantitative method in evaluating the subjective decision-making process in the QFD analysis. Another major drawback of QFD is the need to deal with large amounts of data on a rather subjective basis. A machine-learning approach, using ANN, has been suggested to resolve part of this problem by computing the customer's satisfaction index objectively instead of the customer subjectively ranking the competitors' and in-house products in the customer evaluation part of the house of quality. The Taguchi method has been proposed to help benchmark in the house of quality. QFD identifies the direction of improvement for certain design parameters, but cannot give the exact amount of improvements or the exact target values. The Taguchi method will help determine precise target values for the manufacturing process. The article has reviewed work carried out by authors to integrate fuzzy logic and QFD, neural networks and QFD and Taguchi and QFD, but they were all considered in isolation. Future work will look at how these techniques can be combined together to produce an intelligent systems approach to QFD. A proposed architecture is given in Figure 5. This architecture is intended to take into account the difficulties in correlating two different functions, such as the customer demands and the engineering characteristics. The combination of these methods will help QFD teams move away from thinking that building the house of quality is the only part of QFD. It will help with the propagation of the analysis through the other phases of the QFD process. The merging of these techniques is envisaged to make the QFD process more robust, more quantitatively-oriented, and bring together the different stages of the QFD process. Artificial Neural Network BIJ 7,1 IV Production Important characteristics References Akao, Y. (1972), ``New product development and quality assurance: system of QFD, standardisation and quality control'', Japan Standards Association, Vol. 25 No. 4, pp. 9-14. Akao, Y. (1990), QFD: Integrating Customer Requirements into Product Design, Productivity Press, Cambridge, MA. Chen, S. and Hwang (1992), ``Ranking fuzzy numbers with maximising set and minimising set'', Fuzzy Sets and Systems, Vol. 20 No. 2, pp. 147-62. Clausing, D. and Pugh, S. (1991), ``Enhanced quality function deployment'', Proceedings of the Design Productivity International Conference, Massachusetts, pp. 15-25. Hammerstrom, D. (1993), ``Neural networks at work'', IEEE Spectrum Computer Applications, pp. 26-32. Khoo, L.P. and Ho, C.N. (1996), ``Framework of a fuzzy quality function deployment system'', International Journal of Production Research, Vol. 34 No. 2, pp. 299-311. Masud, A.S. and Dean, E.B. (1993), ``Using fuzzy sets in quality function deployment'', Proceedings of the 2nd Industrial Engineering Research Conference, California, pp. 270-4. Re Velle, J.B. (1991), ``Using QFD with dynamic customer requirements'', GOAL/QPC Research Report, Massachusetts, pp. 10-33. Taguchi, G. (1986), Introduction to Quality Engineering: Design Quality into Products and Processes, Asian Productivity Organisation, Hong Kong. Taguchi, G. (1993), Taguchi on Robust Technology Development: Bringing Quality Upstream, ASME Press. Terninko, J. (1992), ``Synergy of Taguchi's philosophy with next generation QFD'', Transactions of the 4th Symposium on QFD, Novi, pp. 303-15. Urban, G. (1980), Design and Marketing of New Products, Prentice-Hall, Englewood Cliffs, NJ. Wasserman, G., Mohanty, G., Sudjianto, A. and Sanrow, C. (1993), ``Using fuzzy set to derive an overall customer satisfaction index'', Transactions of the 5th Symposium on QFD, Novi, pp. 36-54. Zadeh, L. (1965), ``Fuzzy sets'', Information and Control, Vol. 8, pp. 338-53. Zhang, X.P., Bode, J. and Ren, S.J. (1996), ``Neural networks in quality function deployment'', Computers and Industrial Engineering, Vol. 31 Nos 3/4, pp. 669-73. View publication stats Quality function deployment 19