Six sigma or design for six sigma?

Ricardo Ban uelas and Jiju Antony

Introduction
Six sigma has been considered as a philosophy that employs a well-structured continuous improvement methodology to reduce process variability and drive out waste within the business processes using statistical tools and techniques (Banuelas and Antony, 2002). The main focus of six sigma is to reduce potential variability from processes and products by using either a continuous improvement methodology or a design/redesign approach known as design for six sigma (DFSS). The former follows the phases: dene, measure, analyse, improve and control. This approach is known as DMAIC methodology and is employed in existing processes/products. On the other hand, design for six sigma employs the IDOV (identify, design, optimise and verify) methodology, during the design/redesign of processes or products (Antony and Banuelas, 2002). According to Harry and Schroeder (2000), organisations which have adopted the principles and concepts of the six sigma methodology have realised that once they have achieved ve sigma quality levels (i.e. 233 defects per million opportunities) the only way to surpass the ve sigma quality level is to redesign their products, processes and services by means of DFSS. Although this afrmation is highly arguable, because of lack of data to support the claim and the absence of assumptions used to formulate it, other authors have tended to support it (Chowdhury, 2001; Tennant, 2001). It is not clear if Harry and Schroeders criterion is just applicable to electronic manufacturing processes (e.g. Motorola Company), where much of their work was carried out, or whether it can be applicable to any industry. Moreover, the role of variables such as risk, complexity, new technology, time, cost and customer demands, which may determine the redesign efforts, are not specied. The objectives of this paper are to identify when one approach becomes a priority over the other and to test the applicability of multicriteria decision technique, such as analytic hierarchy process (AHP), to determine the use of DFSS over six sigma methodology. Using action research as a method of inquiry, one six sigma project and one design for six sigma project were carried out. The in-depth participation and action that action
This research has been founded by the Mexican Council of Science and Technology (CONACYT). Special thanks to Martin Brace, Noel Byrne, Phil Rowe, Nick Shubotham, Alwyn Hines, John Powell and Rod Dewey from the collaborative companies. The authors would like to thank the referees whose suggestions helped in the improvement of the nal version of the manuscript.

The authors
Ricardo Banuelas is a Research Engineer at the Warwick Manufacturing Group, School of Engineering, University of Warwick, Coventry, UK. Jiju Antony is Director of the Six Sigma Research Centre, Division of Management, Caledonian Business School, Glasgow Caledonian University, Glasgow, UK.

Keywords
Quality programmes, Analytical hierarchy process, Research

Abstract
Nowadays there is a wider acceptance, growth and investment of six sigma and design for six sigma (DFSS) in both manufacturing and service industries. Although initially employed to improve existing processes, six sigma principles are also applied to design and redesign processes to ensure achievement of high levels of quality. This effort is known as design for six sigma (DFSS). This paper tests the suitability of a multicriteria decision-making technique, the analytic hierarchy process, to make a choice between six sigma and DFSS in two multi-national companies.

Electronic access
The Emerald Research Register for this journal is available at www.emeraldinsight.com/researchregister The current issue and full text archive of this journal is available at www.emeraldinsight.com/0954-478X.htm

The TQM Magazine Volume 16 Number 4 2004 pp. 250-263 q Emerald Group Publishing Limited ISSN 0954-478X DOI 10.1108/09544780410541909

250

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

research provides (Susman and Evered, 1978; Westbrook, 1994; Coughlan and Coghlan, 2002), gives the researchers the opportunity to identify several factors affecting the decision of choosing DFSS/six sigma. It also allows the researchers to legitimately intervene the six sigma practice at the collaborative companies to test the applicability of AHP to deal with this problem.

Background to the research

Six sigma improvement projects work within the framework of a companys existing processes and seek to enhance them by continuous incremental improvements. The aim is to do what the company already does, but doing it efciently by reducing variability from processes. These projects are developed from todays perspective and constrained by assumptions made during development and design stages. Generally improvement projects assume that (Nave, 2002): . the design of product or service is essentially correct and the most economical; . customer needs are satised with that design; and . the current product conguration satises the functional requirements of the market and customer. Divergently, design for six sigma efforts tend to predict and improve quality before products and processes are launched by using the IDOV methodology. It can also be employed to redesign current processes and products. This approach can be utilised to enhance process and product effectiveness[1], and not just efciency[2] and avoid future problems at manufacturing and service stages. IDOV intends to create processes and products that are (Harry and Schroeder, 2000): . resource efcient; . capable of reaching very high yields, regardless of complexity and volume (more than ve sigma); . robust to process variability; and . highly linked to customer demands. The subject of the coexistence of continuous improvement (DMAIC) as an incremental improvement method, and continuous innovation (IDOV) as a radical redesign strategy is not a new subject of discussion (Weston, 2001). Many of the distinctions between DMAIC and IDOV have parallels in other elds. In the quality arena the examples are business process re-engineering in opposition to Kaizen (Hammer and Champy, 2001; Imai, 1986); continuous improvement and continuous innovation (Cole, 2001), in theory of

change examples include rst order change and second order change (Watzlawick et al., 1974); in learning, single-loop learning and double-loop learning (Argyris, 1991). In the vast majority of these cases, scholars conclude that the question is not about the compatibleness between each approach, but the question seems to be regarding the suitability under specic environment conditions of one philosophy over the other (Banuelas and Antony, 2003). Under the six sigma arena, scholars and consultants have tried to set criteria in which one six sigma methodology becomes a priority over the other for current processes and products. Current literature presents two main perspectives about this issue. One of them considers organisations that adopted the principles and concepts of six sigma methodology and have realised that once they have achieved ve sigma quality levels (i.e. 233 defects per million opportunities) employing DMAIC improvement methodology, the only way to surpass the ve sigma quality level barrier is to redesign their products and processes from scratch by means of DFSS (Harry and Schroeder, 2000; Chowdhury, 2001; and Tennant, 2001). The other hypothesis states that this decision is not straightforward, given the fact that many variables need to be taken into account, consequently just a general guideline can be provided. Such guideline is based on a number of criteria such as risk, process capability, technology availability and changes in companys strategy (Pande et al., 2000; Eckes, 2001) However, both hypotheses present some assumptions that open the discussion. First, the few variables presented in the work of Pande et al. (2000) and Eckes (2001) just produce a general guideline since these variables are excessively broad and leave minuscule opportunities to evaluate them. In addition, they do not introduce analytical tools and procedures that might be used in this context. When very general variables are set, such as those of Pande et al. (2000) and Eckes (2001), people make decisions by satisfying one variable/criterion at a time. They focus upon options that satisfy the most important criterion and eliminate the remaining options. This may leave the whole organisation more disturbed than it was before (Saaty, 1988). Second, simplistic recommendations in which just one variable is considered (e.g. DFSS need to be used when processes reach ve sigma level (Harry and Schroeder, 2000; Chowdhury, 2001; Tennant, 2001)) could lead to the selection of improvement on areas using an erroneous approach. Moreover, the sigma level metric is constructed upon not technical rigorous approach, such as the assumption that all processes shift by 1.5

251

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

sigma from the process mean and the ambiguity in setting the number of critical to quality characteristics that may modify the sigma level. Nowadays there is a wider acceptance, growth and investment of six sigma in industry (Linderman et al., 2002). Thus, the potential value of management technique to manage the six sigma approach selection could lead organisations applying six sigma to focus their efforts using the appropriate approach. During the course of this paper, it is argued that it is useless to try improving processes when their fundamental design is wrong. Therefore, it is desirable to identify conditions where the processes effectiveness can be improved by employing redesign efforts. This will also facilitate companies to make an appropriate resource allocation when employing the most suitable six sigma approach. Consequently, efforts focused upon increasing customer satisfaction and business performance will be correctly addressed. To deal with this issue the researchers present two action research projects; one six sigma project and one design for six sigma project, in which they facilitated company efforts to achieve high quality standards playing the role of black belts.

stages. It is performed in a cyclical process which aims to increase understanding of a given situation (Baskerville and Wood-Harper, 1996). This cyclical process is not a random method; it is systematic, involving a self-reective spiral or cycle of planning, acting and fact nding through self-reecting (McNiff, 1998). This research starts by establishing the theoretical foundation of the research and selecting a research design. After that, two core action research cycles were carried out which follow the planning, action and reecting phases Rigorous reection about the actions, methodologies and the underlying assumptions and perspectives made this reection to an innovative application of knowledge and gives the appropriate research reliability (Coughlan and Coghlan, 2002).

Core action research: the rst cycle at collaborative company A

Planning The rst action research project took place at one of the UK subsidiaries of a multinational company recently embarked on six sigma. This company manufactures and designs a wide range of abrasives and focuses its six sigma efforts on waste reduction. These efforts are linked to the business critical UK regional Y[3] of reducing manufacturing costs. Six sigma DMAIC approach was used to reduce waste in one of the manufacturing lines. This line is a continuous process with equipment designed to allow a non-stop production during roll changes material and the unloading of nished rolls of production (Shepherd, 1995). It is not fully automated and operators are necessary to unload nished rolls and load new rolls of web material coming from previous production process. The re-winder machine at the end of the line allows the line to run continuously. However it frequently fails to change from one roll to the other. A failed chop-over results in a loss of web tension and therefore a line stop. The purpose of this six sigma project was to identify, quantify and eliminate the source of variation that results in failure to change from one spindle or roll to another by the re-winder machine. As most of the six sigma DMAIC projects, the re-winder project works within the framework of a companys existing processes. This project is developed from todays perspective and constrained by assumptions made during development and design of the re-winder machine. Therefore it was initially assumed that this design is essentially correct and the most economical;

Research methodology
The research methodology was developed primarily by action research. Hult and Lennung (1980) meticulously dene action research as:
. . . a research strategy which simultaneously assist in practical problem solving and expand scientic knowledge, as well actors, being performed in immediate situation using data feedback in a cyclical process aiming at an increasing understanding of a given social situation, primary applicable for the understanding of change processes in social systems and undertaking within mutually acceptable risk.

The assumptions in which action research is based are placed within the phenomenological paradigm (Hussey and Hussey, 1997). The study of new or changed techniques and methodologies implicitly involves the introduction of such changes, and it is necessarily interventionist. From the collaborative companies viewpoint, the study of a technique is impossible without intervening in some way to inject the new technique into the companys environment (Coghlan and Brannick, 2001). Action research is one of the few valid research approaches that can be legitimately employed to study the effects of specic alterations in systems development methodologies in organisations (Baskerville and Pries-Hege, 1999). Action research is an organic process involving systematic and sometimes interactive

252

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

delivery partners needs were satised with that design and it is congured to satisfy the functional requirements of the manufacturing line.

Action During the action phase interventions are made, that is, DMAIC project is carried out to study how DMAIC became a priority over DFSS and to test the applicability of multicriteria decision technique, such as analytic hierarchy process (AHP). The rst step of the DMAIC methodology involves identifying the projects critical to quality/cost characteristics, approving team charter and estimating nancial benets (as shown in Table I). Thus, a team charter was carried out to state the opportunity that exists in nancial terms. It summarises the dene stage from the business critical Y and its linkage to the project Y. It also cascades down the project description, goals and potential nancial benets. Having dened the project, the team moved to the measure phase. The measurement phase has the purpose of dening the current process and establish metrics that describe the project Y performance in order to narrow the problem to its major factors or vital few root causes (Pande et al., 2000). To this end, the process was represented graphically using standard operation procedures. Using rational sub-grouping the process baseline was calculated. The initial process capability was 1.2 sigma longterm (88.5 per cent yield) and included a 1.2 sigma shift. A data collection plan was formulated in which potential Xs were recorded during a relative long period of time. The data collected
Table I DMAIC tool usage Step Specic tasks 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Project scope and boundary Dene defects Team charter and team leader Estimated nancial impact Leadership approval

from the measure phase served as an input for the analysis phase. The purpose of the analyse phase is to start learning about data in order to generate, segment, prioritise, and verify the possible root causes and its relationship to the Ys or outputs (Waddick, 2001). Different tools and techniques were employed to relate the Y to the Xs in order to reduce the number of factors and select the vital few for further analysis (as shown in Table I). For example, main effect plots, multi-vari charts, hypothesis tests, descriptive statistics, normality tests and graphical representations were conducted. Multi-vari study played an important role during this phase. Multi-vari studies helped to associate the Y and the key Xs, identify noise variables and reduce the Xs for the improvement phase. In addition, it was helpful to obtain an in-depth understanding of the process during its natural variation. In fact, during the multi-vari study some special causes were eliminated and countermeasures were implemented as the process was understood. The improvement phase has the objective to consider the causes found in the analysis phase, and also select and target solutions to eliminate such causes in order to achieve the improvement goals set during the dene phase. From the multi-vari study the team understood the relationship among different factors and identied the source of variation and its impact on the rewinder. The team developed an improved system to reduce the variability of the main critical to quality characteristic. This system allows operators to know the optimal value of the critical to quality characteristic and to set the re-winder in that

Focus Y Y Y Y Y Y, X Y, X Y Y Y, X Y, X X X Y, X Y, X Y Y, X

Tools and techniques employed Team charter; Pareto; critical to cost tree; sub-process map; process map (SIPOC); brainstorming; run charts; cause and effect analysis Cause and effect matrix; FMEA; gauge R&R; process capability; dpmo calculation; run charts; box plots Capability analysis; multi-vari study; hypothesis testing; graphical analysis

Dene

Measure

Map process and identify inputs and outputs Cause and effects matrix Establish measurement system capability Establish baseline capability of the process Complete FMEA Perform multi-vari analysis Identity potential critical inputs Develop plan for next phase Verify critical inputs Optimise critical inputs Implement control plan Verify long-term capability Continuous process improvement

Analyse

Improve

Statistical tolerancing; analytic hierarchy process; cost/benet analysis; ANOVA; risk assessment; rational sub-grouping Response plan; control plans; procedures; mistake proong operation; standard operation procedure

Control

253

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

optimal setting. This improved method allows improving the process to 2.7 sigma level. However, through the knowledge obtained from the dene, measure, analyse and improve phases the team suggests a redesigned alternative, which involves the reengineering of the turret indexising system of the re-winder, which potentially reduce the variability radically. However, there are other factors such as risk, the potential savings and the costs involved in which the redesign of the re-winder is not necessarily the most suitable option. Making a good decision involves making trade-offs between multiple objectives to select an alternative that best meets the values of the decision maker. The authors, with the support of the six sigma team at the company, decided to employ analytic hierarchy process (AHP) to determine the suitable use of redesign alternative over the improvement option. AHP usage on a wider range of scenarios and accuracy is widely supported by the literature (Saaty, 1988; Banuelas and Antony, 2003). AHP is a technique for solving decision problems using the following ve steps (Rosenbloom, 1996): (1) Create a decision hierarchy by breaking down the problem into a hierarchy of decision elements. (2) Collect input by a pairwise comparison of decision elements. (3) Calculate the relative weights of the decision elements. (4) Determine whether the input data satises a consistency index. If does not, go back to step 2 and re-do the pairwise comparisons. (5) Aggregate the relative weights to obtain the scores and hence rankings for the decision alternatives. Figure 1 depicts the AHP process in ow chart format. For more information on the subject the reader is referred to Saaty (1988; 1996) and Banuelas and Antony (2003). AHP was carried out, having as main objective the selection of the re-winder redesign over the current improvement. This problem was decomposed into the decision elements shown in Figure 2 (i.e. maximise nancial benets, maximise process capability, cash avoidance and reduce risk). In order to determine the relative importance among these objectives at each level, AHP measures strength of importance by pairwise comparisons and puts the results into a matrix. This pairwise comparison was performed during brainstorming sessions where the members of the six sigma team participated. The scale value proposed by Saaty (1988) was used to make pairwise comparisons based on judgments (see Table II).

Figure 1 AHP ow chart

From the scale in Table II, the following pairwise comparison matrix A shown in Table III was calculated (Step 2 in Figure 1). Table III indicates how much more important is the ith objective than the jth objective and this will lead to the construction of the column vector of objectives importance. For example, in this case the team decided that the objective Maximise nancial benets is slightly more important than Maximise process capability; therefore, based on Table II a number 3 was assigned. Hence, reciprocally the objective Maximise process capability is 0.33

254

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Figure 2 AHP in a six sigma project

Table II Scale of preferences between two elements Value of aij 1 3 5 7 9 2,4,6,8 Interpretation Equal importance Weak importance of one over the other Essential or strong importance Very strong or demonstrated Absolute importance Intermediate values between adjacent scale values

Source: Saaty (1988)

(1/3) times less important than the objective Maximise nancial benets producing with this a quadratic reciprocal matrix, A. After the A matrix was calculated, the priority vector of objectives was calculated. The rst step is to calculate the c vector in order to estimate the normalised matrix Aw. Aw is produced by dividing each entry in column i of A by the sum of the
Table III Pairwise comparison matrix A No. A= 1 2 3 4 Objective description Maximise nancial benets Maximise process capability Cash avoidance Minimise risk Maximise nancial benets 1.00 0.33 0.33 1.00

entries in column i (Saaty, 1988; Chuang, 2001). Finally, by computing the ci as the average of the entries in row i of Aw the column vector c (Table IV) is produced; where c represents the relative degree of importance for the ith objective. The resultant vector produces the priority weightings according to the pairwise comparison carried out. For each level of the hierarchy the same procedure is carried out (A, Aw and c calculations). In order to produce an overall rating for each alternative, AHP combines the priority weightings of objectives and attributes with the comparison rating for alternatives by multiplying the priority weight of each objective i by the comparison for alternative j with respect to objective i. As a result, in this case the redesign alternative may satisfy the selected objectives greater than the improved version of the process (0.6412 vs 0.3588). Based on this result, the six sigma team decided to redesign the process using new drives, an encoder and brake disk to reduce

Objectives Maximise process capability 3.00 1.00 1.00 2.00

Cash avoidance 3.00 1.00 1.00 3.00

Minimise risk 1.00 0.50 0.33 1.00

255

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Table IV Priority weightings No. C= 1 2 3 4 Sub-objective description Maximise nancial benets Maximise process capability Cash avoidance Minimise risk Priority weight 0.3829 0.1423 0.1276 0.3472

the variability of the critical to quality characteristic. Having identied the root causes of the problem and the possible solution to reduce the variation of the process, the six sigma team moved to the control phase. This phase has the objective of implementing ongoing measures and actions to sustain the improvement by monitoring, standardising, documenting and integrating the new process into daily activities (Pande et al., 2000). The control method consists of a standard operation procedure (SOP) with reaction plan. The SOP document describes how to perform the operation. In addition, it describes the reaction plan in case of failure. The solution is considered by itself a mistake proong solution. The saving of this project is estimated to be 100,000 and reached 3.3 sigma level long term.

Fact nding The evidence in the form of data and backing information of this project indicates the capability of the process as a positive association with the selection of redesign or improvement efforts within six sigma. Analogous to the six sigma literature, which suggest that six sigma methodology is capable to improve process capability as the cause of variation is understood, through the course of the project different process capabilities were reached as the root causes were understood and eliminated using DMAIC, as shown in Figure 3. However, it contrasts with the literature, which suggests that the only way to surpass the ve sigma wall is by redesigning the process. In this case the redesign option was employed to surpass the 2.7 sigma wall. In addition, using rational sub-grouping to estimate sigma capability level, it was possible to calculate the real process shift and discard the 1.5 sigma shift. Nevertheless, the sigma level was not the only variable that was taken into account to select redesign effort over the improvement alternative. The six sigma team identied nancial benet, cash avoidance, risk and capability as the main factors affecting this decision. These factors were further sub-divided into attributes. Financial aspect was a fundamental variable that affected the decision regarding when to embark on a redesign or improvement option. The belief that redesign requires larger investment than continuous

improvements coincides with this project. The redesign option investment is about 50 times more costly than that of the improvement costs needed to improve the process from 1.8 sigma to 2.7 sigma. However, the amount of waste and the possibility to automate the operation led a rapid return on investment. It is a general belief that redesign of a process may be a more risky approach than continuous improvement (Leach, 1996; Pande et al., 2000). However, it was anticipated that project resulted in high risk if the continuous improvement approach is chosen over the redesign alternative. The rapid change in the current manufacturing technology is making the support of it both risky and expensive. The spares for the current machinery are becoming obsolete, increasing their acquisition costs and the lead-time to supply the spares. In addition, the ageing of its components means the likelihood of breakdown and inaccuracy greatly increases the older this technology becomes. All this comes to a point where apparently redesigning the system is a more feasible solution than trying to improve the existing technology. The objective cash avoidance presented a trade-off between the two alternatives. The redesign alternative requires a substantial initial investment, whereas the improvement alternative does not require any initial investment. Nevertheless, the redesign option is capable of reducing the annual costs by approximately 100,000, whereas the improvement alternative by approximately 50,000 annually. There were many factors involved when considering redesign efforts instead of continuous improvement; from the scope and capability of the process, to the possibility of waste reduction avoiding major expenditure, to the willingness to undergo a business upheaval, to the urgency of major performance gains (Pande et al., 2000). These sorts of decisions during six sigma projects at the collaborative company are highly based on the experience of people and decided without the application of tools and techniques. However, this six sigma project employed the AHP to deal with this multicriteria decision. There can be little doubt about the positive implementation of AHP in the six sigma arena. The realisation, understanding and awareness that involved the comparison of alternatives, so making it necessary to consider them, was, in itself an advance. The awareness that is necessary to consider the different objectives that may affect the situation made the decision makers contemplate the outcomes of the decision making much more carefully. It is important to mention that the tools, techniques and methods employed on this

256

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Figure 3 Process improvement trend (in sigma terms)

DMAIC project do not differ signicantly, to those employed on the DFSS. An examination of the black belt training material at the collaborative company produces signicant similarities between DMAIC and DFSS regarding the tools taught. During this DMAIC project most of the DMAIC tools and techniques were employed. However, not all the tools employed in this project are part of the DMAIC toolkit. Statistical tolerancing, a DFSS technique, was employed during the improvement phase. During the course of this project the use of DMAIC seemed to be clear because the approach to tackle it was highly defensive and it was assumed that the design was correct. However, it was not until the root cause of variability is understood that the six sigma team proposed a redesign alternative to reduce process variation. In addition, the redesign alternative was not suggested by employing DFSS but by employing DMAIC and understanding the root causation and the possible way to eliminate it. Therefore, in some instances, the selection of DFSS over DMAIC may take place during the course of either a continuous improvement DMAIC project or during a DFSS project. Another issue where DFSS and DMAIC traditionally diverge, is the scale of improvement associated with both approaches. Traditionally, DFSS as any redesign effort, is associated with larger and discontinuous improvement, whereas DMAIC is usually associated with steady continuous improvement. During this project, the improvement efforts can be associated with the slope and redesign and can be classied as a radical

improvement, as a staircase in Imais words. However, their scale of improvement reached using DMAIC compared with that expected by redesigning the re-winder machine do not necessary differ in scale. Therefore this distinction does not necessary apply to this specic project. According to Cole (2001) continuous improvement and innovation do not necessarily need to differ in scale and there is no logical reason to associate the term innovation (or redesign) with large-scale discontinuous change. Cole (2001) argues that there is plenty of innovation that occurs in the course of continuous improvement.

Core action research: the second cycle at collaborative company B

Planning On the DMAIC project presented above was shown how DMAIC and DFSS play complementary roles, however, it just represents one part of the spectrum. Therefore the researchers were primary concerned with the study of a DFSS project in order to be more able to generalise conclusions. This DFSS project was carried out in the home laundry appliance division of one of the largest suppliers of white goods in Europe. Nowadays, in the whites goods industry, there is a growing awareness among consumers and governments of the need to conserve both energy and water. Energy efciency labels and standards

257

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

for appliances and equipment are playing key roles in the governments strategies to meet energy and environmental goals. These labels and standards are widely used to improve and measure the efciency of home appliances. Accordingly, under the European Union direction, A rating represents the highest rating awarded for a model for any or all of these features (IEA, 2000). The second action research cycle was launched aimed to achieve B rated energy consumption (i.e. the second best energy rating for a domestic appliance) for the new platform of one of the companys products using DFSS. Action The rst step in DFSS involved the gathering of the voice of the customer and translating it to critical to quality (CTQ) characteristics. In this project the CTQ under study is energy performance of one of the domestic appliances produced by the company. The energy consumption of the product is dictated by several engineering sub-systems or elements. This project is mainly concerned on the selection, designing and optimising of the heat element of the appliance in order to facilitate the achievement of B energy in the nal product. To achieve this primarily goal, it is necessary to select the best design concept from which to proceed to detailed design optimisation and ultimately to manufacture. This selection involves comparing the current design against a set of heat elements currently in use by competitive products. Therefore, if the current heat element is selected over the rest, then the team of designers will proceed to improve its energy efciency. However, if a different heat element concept is selected, the team will proceed to detail redesign of the heat element and the sub-systems affected by a new heat element using DFSS. In addition, the selection of the most appropriate heat element for the new platform of the appliance not only depends on the aptness of the heat element to facilitate B energy level but also an improved design for manufacturing score, low cost and reliability targets. Without applying concept design selection tools and techniques, it is extremely easy to select the inappropriate concept based on simplistic solutions and not the one which maximises the utility of different, and sometimes, conicting design parameters. In order to assist with the selection made on concept design selection, AHP was chosen and presented to the collaborative company by the researchers. Following the ow chart presented in Figure 1, subject matter experts performed the analytic hierarchy process forming the hierarchy shown in Figure 4. As a result, the ranking in

Table V was calculated using the standard AHP procedure. Sometimes a clear apparent winner among alternatives will emerge using AHP. However, in this case Concept Design #3 and #2 are very close to each other (0.2954 vs. 0.2942). For Scott (2002) the nal rankings are used as a general guide in the selection of a particular alternative, but a small difference in scores is not be taken as denitive evidence that one alternative is preferable to another. To provide a mean of testing statistical difference in the rankings, the researchers modied the basis of the AHP as follows: (1) Create a decision hierarchy by breaking down the problem into a hierarchy of decision elements. (2) Collect statistical input by pairwise comparisons of decision elements forming statistical distribution of preferences. (3) Calculate the relative statistical weights of the decision elements. (4) Determine whether the input data satises a consistency index. If it does not, go back to step 2 and redo the pairwise comparisons. (5) Aggregate the relative weights using Monte Carlo simulation to obtain the statistical distribution of the scores. (6) Determine the statistical difference of the aggregated relative weights and hence rankings for the decision alternatives. This modied approach, called stochastic AHP (SAHP), employs probabilistic distributions instead of using a deterministic scale as shown in Figure 5. For example, the ith objective can be preferred over jth with an average of x, between 1-9, and standard deviation of z. That is, the A matrix of the AHP contains probabilistic distributions (e.g. normal, triangular or uniform distributions). Monte Carlo simulation is then used to calculate the Aw and c vector. Accordingly, each replication would be a realisation of all the ai,js in the decision hierarchy followed by the standard AHP calculation. Replicating n times will provide estimates of the probabilities associated with the vector of priorities (Rosenbloom, 1996). Based on this principle, the team created a SAHP model shown in Figure 5. This model puts special emphasis on the inputs, such as problem recognition and formulation, subject matter experts selection, scope and boundaries of the AHP. The resultant vector of priorities, represented in Figure 6 is treated using statistical tools and techniques in order to obtain statistical signicance of the results. Figure 6 can be viewed as providing a measure of statistic signicant differences between the alternatives that can lead to prioritising order of alternatives.

258

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Figure 4 AHP in DFSS project

Table V The deterministic c vector No. C= 1 2 3 4 Alternatives Current design concept Concept design 1 Concept design 2 Concept design 3 Priority weight 0.2078 10.2026 0.2942 0.2954

Using one way ANOVA, it is then possible to investigate the question: Are the means in the c vector (ci) different? In this case the means of the ci are considered signicantly different. Thus, a new concept was selected over the current design concept and two design concepts used in competitive products. Since the current design was not selected, a major redesign took place in which different design parameters were optimised. Nevertheless, if the current design concept will be selected after SAHP, the team will proceed to improve the current design by means of IDOV methodology. In other words, during a design project it is possible to improve the design by minor improvement or optimisations. After design concept was selected, the team move on to the optimisation phase of the IDOV methodology (see Table VI). The optimise phase has the objective of optimising the selected design concept and makes trade-off between competing requirements. Screening experiment was employed to discover the critical few Xs that affect the heat element performance. Having identied the critical parameters, response surface methodology (RSM) was carried out to determine the settings of the design parameters that produce the optimal response(s) (Khuri and Cornell, 1996). Multiple

response optimiser was then employed to assist with the identication of the combination of parameter settings that jointly optimise two responses (i.e. time, energy consumption). The optimised redesigned heat element was then validated to conrm that the pilot experiments build the predictions made during the RSM. As a result, the product achieved half energy band improvement. It is expected that other parallel projects to this one will provide the extra energy consumption reduction to achieve the critical to business Y: B energy.

Fact nding During this DFSS project the sigma capability criteria to dene the use of a redesign approach was not employed. The change of the current product specication to a higher level left the inherent technology incapable of reaching the new customer demands. Pande et al. (2000) and Eckes (2001) propose that process/product redesign is suitable when a new process will assist an organisation to achieve a strategy objective, as is the case of this DFSS project. However, similar to the DMAIC project, this project was redesigned based on multicriteria rather than the sigma level alone. Risk, cost and cash avoidance played again an important role, along with improved product safety, reliability, performance and the ability to be recycled. The application of the traditional AHP, to deal with this multicriteria decision presented some drawbacks in this project. In order to deal with the uncertainty in the judgments needed to construct the AHP and to provide a mean for statistical

259

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Figure 5 The stochastic AHP model

signicance in the alternatives, SAHP was developed. Although this method is more complex that the original AHP and other multi-criteria decisions models, the subject matter experts do not require advanced knowledge of either mathematics, statistics or AHP to construct the hierarchy and perform the pairwise comparisons. They fundamentally need to focus on the input of the model Figure 5 and soft issues. The rest of

the model can be carried out using Crystal Ball software. The management of uncertainty using probabilistic judgments provides to the SAHP a mean to test the nal scores and makes it compatible with the design for six sigma philosophy, in which variability plays an important role in the decision. With the use of this modied approach the team was able to identify and quantify the multicriteria in this decision.

260

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Figure 6 Probabilistic vector of priorities

Table VI DFSS tool usage Specify tasks Tools employed Team charter; critical to quality tree; quality function deployment; benchmarking; CTQ ow down

Identify

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

Design

Optimise

Verify

Identify customer CTQs Perform CTQ ow down Analyse measurement system capability Generate validate systems/subsystems models Build scorecards prediction, parts, process and peformance Roll-up scorecards for all subsystems Capability ow-up Identify the gaps Low Zst on scorecard Lack of transfer function Unknown process capability Use DOE to nd transfer function and the critical few Xs Parameter and tolerance design Generate purchase and manufacturing specications Specify mean, variance and limits Conrm that pilot studies match predictions Mistake proong process Develop manufacturing and supplier control plans for CTQ Rene models, scorecards, process characterisation database

Analytic hierarchy process; regression analysis, baselining, rational subgroups; design for assembly

Full factorial DOE; response surface methodology; multiple response optimiser

Baselining; rational sub-groups

Although the project was initially focused on redesign of the current heat element technology, the decision whether improve or redesign was not answered after different design concepts were evaluated by means of SAHP. Four different design alternatives, including the current design, were evaluated. Since the current design was not selected, a major redesign took place in which different design parameters were optimised. Nevertheless, if the current design concept was selected after SAHP, the team would have proceeded to improve the current design by means of DMAIC. The tools, techniques and methods

employed on this IDOV project do not differ signicant to those employed on the DMAIC. An examination of the six sigma and design for six sigma training materials at the second collaborative company also produced signicant similarities. Not all the tools employed in this IDOV project are exclusive of the DFSS toolkit. Team charter, design of experiments, response surface, critical to quality tree, baselining, rational sub-grouping, multiple response optimiser and AHP are example of tools that can be employed in both methodologies. As shown above, in some cases the selection of DFSS over DMAIC may take

261

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

place during the course of either a continuous improvement DMAIC project or during a DFSS project. However, it is important to note that DFSS is not six sigma for the design department. DFSS has a different approach to quality improvement than DMAIC projects.

Notes
1 Effectiveness is dened as the organisations ability to maximise its returns by whatever means, including not only efciency but also management of its input and output environments (Roberts and Hunt, 1991). 2 Efciency is dened as the technical ability of the organisation to minimise the cost of transforming specied inputs into acceptable outputs (Roberts and Hunt, 1991). 3 The basic equation of six sigma, Y f(x), denes the relationship between a dependent variable (Y) or outcomes of a process and independent variables (the x) or possible causes of problems associated with the process (Brue and Launsby, 2003).

Conclusion and directions for further research

Six sigma has been embraced in many companies as the new business strategy. It is focused on improving processes by eliminating variation using a well-structured methodology. Although initially employed to improve existing processes, six sigma has started to focus upon design and redesign processes capable of achieving six sigma levels using a different methodology. The main objectives of this paper were to identify when one approach becomes a priority over the other and to test the applicability of multicriteria decision techniques, in particular analytic hierarchy process (AHP), to determine the use of DFSS over six sigma DMAIC methodology. Two projects in two different industry sectors have been examined using action research as a method of investigation. From the research, it can be concluded that the decision as to when to employ a redesign approach in a six sigma project is characterised by the following issues: . the ve sigma wall is not necessarily the only and most accurate criterion to select between six sigma and DFSS; . numerous factors affect the decision improve/ redesign decision; . multicriteria decision analysis techniques (e.g. AHP and SAHP) aid with the selection of six sigma and/or DFSS; . six sigma projects can have traces of redesign activities and vice versa; . some tools and techniques apply for both methodologies; and . the decision as to when to embark on a redesign activity can occur at different stages of the DFSS and DMAIC methodologies (e.g. project selection and concept selection). As with any research effort, there are additional areas which need to be investigated and improved upon. The next stage of this research is to study the improve/redesign question in a six sigma transactional project. This will enable the testing of SAHP under different conditions and identify other factors affecting this selection process.

References
Antony, J. and Banuelas, R. (2002), Design for six sigma, IEE Manufacturing Engineering, Vol. 81 No. 1, pp. 119-21. Argyris, C. (1991), Teaching smart people how to learn, Harvard Business Review, May-June, pp. 99-109. Banuelas, R. and Antony, J. (2002), Critical success factors for the successful implementation of six sigma projects in organisations, The TQM Magazine, Vol. 14 No. 2, pp. 92-9. Banuelas, R. and Antony, J. (2003), Going from six sigma to design for six sigma: an exploratory study using AHP, The TQM Magazine, Vol. 15 No. 5, pp. 334-44. Baskerville, R. and Pries-Heje, J. (1999), Grounded action research: a method for understanding IT in practice, Accounting Management and Information Technologies, Vol. 9 No. 1, pp. 1-23. Baskerville, R. and Wood-Harper, T. (1996), A critical perspective on action research as a method for information systems research, Journal of Information Technology, Vol. 11 No. 3, pp. 235-46. Brue, G. and Launsby, R. (2003), Design for Six Sigma, McGraw-Hill, New York, NY. Chowdhury, S. (2001), The Power of Six Sigma, FT/Prentice-Hall, London. Chuang, P. (2001), Combining the analytic hierarchy process and quality function deployment for a location decision from a requirement perspective, International Journal of Advanced Manufacturing Technology, Vol. 18 No. 11, pp. 842-9. Coghlan, D. and Brannick, T. (2001), Doing Action Research in Your Own Organisation, Sage, London. Cole, R. (2001), From continuous improvement to continuous innovation, Quality Management Journal, Vol. 8 No. 4, available at: www.asq.org/pub/qmj/past/vol8_issue4/ cole.html (accessed 27 March 2002). Coughlan, P. and Coghlan, D. (2002), Action research for operations management, International Journal of Operations & Production Management, Vol. 22 No. 2, pp. 220-40. Eckes, G. (2001), The Six Sigma Revolution, John Wiley & Sons, New York, NY. Hammer, M. and Champy, J. (2001), Reengineering the Corporation, HarperBusiness, New York, NY. Harry, M. and Schroeder, R. (2000), Six Sigma; The Breakthrough Management Strategy Revolutionizing the Worlds Top Corporations, Currency, New York, NY.

262

Six sigma or design for six sigma?

The TQM Magazine Volume 16 Number 4 2004 250-263

Ricardo Banuelas and Jiju Antony

Hult, M. and Lennung, S. (1980), Towards a denition of action research: a note and bibliography, Journal of Management Studies, Vol. 17 No. 2, pp. 241-50. Hussey, J. and Hussey, R. (1997), Business Research: A Practical Guide for Undergraduate and Postgraduate Students, Palgrave, New York, NY. IEA (2000), Energy Levels and Standards, International Energy Agency, Paris. Imai, M. (1986), Kaizen: The Key to Japans Competitive Success, McGraw-Hill, New York, NY. Khuri, A. and Cornell, J. (1996), Response Surfaces. Design and Analysis, Dekker, New York, NY. Leach, L. (1996), TQM, reengineering and the edge of chaos, Quality Progress, Vol. 29 No. 2, pp. 85-90. Linderman, K., Schroeder, R., Zaheer, S. and Choo, A. (2002), Six sigma: a goal-theoretic perspective, Journal of Operations Management, Vol. 21 No. 2, pp. 193-203. McNiff, J. (1998), Action Research: Principles and Practices, Macmillan Education, Basingstoke. Nave, D. (2002), How to compare six sigma, lean and the theory of constraints, Quality Progress, Vol. 35 No. 3, pp. 73-9. Pande, P., Neuman, R. and Cavanagh, R. (2000), The Six Sigma Way: How GE, Motorola and Other Top Companies Are Honing Their Performance, McGraw-Hill, New York, NY. Roberts, K. and Hunt, D. (1991), Organisational Behavior, PWS Publishing Co., Boston, MA. Rosenbloom, E. (1996), A probabilistic interpretation of the nal rankings in AHP, European Journal of Operational Research, Vol. 92 No. 2, pp. 371-8. Saaty, T. (1988), The Analytic Hierarchy Process, Pergamon Press, New York, NY.

Saaty, T. (1996), The Analytic Network Process, RWS Publications, Pittsburgh, PA. Scott, M. (2002), Quantifying certainty in design decisions: examining AHP, Proceedings of the ASME DETC 2002 Design Engineering Technical Conferences: Design Theory and Methodology Conference, Montreal, September, ASME, New York, NY. Shepherd, F. (1995), Modern Coating Technology Systems for Paper, Film and Foil, EMAP Maclaren, Barnet. Susman, G. and Evered, R. (1978), An assessment of the scientic merits of action research, Administrative Science Quarterly, Vol. 23 No. 4, pp. 582-603. Tennant, G. (2001), Six Sigma: SPC and TQM in Manufacturing and Services, Gower, Aldershot. Waddick, P. (2001), Six sigma DMAIC quick reference, available at: www.isixsigma.com (accessed 16 August 2003). Watzlawick, P., Weakland, J. and Fish, R. (1974), Change. Principles for Problem Formulation and Problem Resolution, Norton & Co., New York, NY. Westbrook, R. (1994), Action research: a new paradigm for research in production and operations management, International Journal of Operations & Production Management, Vol. 15 No. 12, pp. 6-20. Weston, F. (2001), The need for continuous improvement and continuous innovation, Quality Management Journal, Vol. 8 No. 4, available at: www.asq.org/pub/qmj/past/ vol8_issue4/qmjv8i4weston.pdf (accessed 27 March 2002).

263