You are on page 1of 8

Comparison of Selected Quality Improvement Methodologies: Shainin, Taguchi and Classical

Bill Ross
Six Sigma Associates

Introduction In the rush toward continuous improvement there is a strong tendency to believe one guru has all the answers. This comes about in large part due to wanting an easy cook book approach which will not impact the current comfort zone of many companies and groups. Unfortunately, many gurus (and their followers) are quick to fill this need. This motivates gurus to not only push their approach, but denigrate others. In reality, the more conversant a company is in each methodology the greater the companys choices to apply those methodologies effectively. This paper will provide a greatly simplified examination of selected elements (tools and techniques) considered part of the philosophies of Dr. Genichi Taguchi and Dorian Shainin. In addition, a simplistic comparison will be made to classical philosophies. applicable, the author will provide assessment of proposed methodologies. Where

Sigma Science Inc.

Ross (303) 494-8521

Comparison of Selected Quality Improvement Methodologies: Shainin, Taguchi and Classical

Shainin Methods Dorian Shainin, a quality consultant, prescribes methodologies that include, but are not limited to, the following:

1.

Sources of variation. The core of Shainins methods is identifying the source of variation. Shainin focuses on

identifying the vital few factors (product or process parameters) that have a significant effect on some desired output. He employs multi-vari studies, component search and variable search techniques to accomplish this. Assessment: Identification of the sources of variation (i.e., cause of the problem) is vital in any methodology. Shainins methods rely heavily on vast engineering knowledge and experience. The multi-vari study is a simple graphical technique to assist in identifying families of variation (i.e., positional, cyclical, temporal), and is a technique that should be used. The component and variable search techniques, however, are suspect: ! Assumptions are questionable (partial list): ! ! ! ! Best and worst levels of a parameter are known. Sample is representative of the total process (inference Relationships that exist are linear. Noise factors are constant through time. space).

Statistical validity only good on samples tested. Efficiency not much better than one-factor-at-a-time experimentation. May isolate parameters, but not lead to improvement. Does not offer a direction to go in for continuous improvement.

2 of 8

Bill Ross (303) 494-8521


Six Sigm a Associates

2. Process control. Shainin believes control charts are useless. He recommends a technique called precontrol to detect product discrepancies (pre-control depends heavily on item 4 discussed herein). Assessment: Pre-control is a part control tool to determine conformance to specification and is not useful in reducing variation from target. It assumes a highly capable (Cpk > 1.5) process that is in control. Pre-control may be useful for a process that is very mature (optimized and in control), but is not recommended otherwise.

3. Process improvement. Shainin uses classical methods of model building. Full factorial experiments are the key tool used (these are classical methods). Assessment: Full factorials have been proven to provide sufficient information for

building linear models complete with interactions. Linear models, however, may not allow us to achieve maximums or optimal conditions. For this classical methods such as EVOP, response surface methods and central composite designs are highly effective.

4. Tolerancing (for both product and process parameters). Identifying appropriate tolerances is another important aspect of Shainin methods as tolerances become critical to decision making. Shainin uses iso-plots and tolerance parallelograms (a type of tolerance correlation study) to determine proper tolerances. Included in the iso-plots methodology is a quantification of measurement system variation. Assessment: Excellent idea to be used by product and process design engineers. Tools are simple to use but effective. However, the measurement system evaluation method is not extensive enough to give substantial enough information for improvement.

3 of 8

Comparison of Selected Quality Improvement Methodologies: Shainin, Taguchi and Classical

Taguchi Methods

1. Sources of variation. Taguchi has been a leader in popularizing the reduction of variation around some target value. He has gone so far as to develop a measurement system, the loss function, to quantify the cost associated with variation from target. Taguchi uses engineering knowledge combined with orthogonal arrays (fractional factorials) to effectively identify sources of variation and then reduce variation. Key in his philosophy is the identification of response variables robust to interactions. Assessment: Taguchis methods, albeit not the most statistically efficient, are effective at identifying sources of variation (main effects) and setting those sources to reduce variation and achieve target in some response variable. A key to his methods is the concept of robust process where his experiments occur in a noisy environment (inner and outer arrays). This may be somewhat complicated, but has proven effective. response variables is an excellent, but poorly understood, idea. Taguchis focus on efficient

2. Process control. Taguchi has been very innovative in his approach to process control. Rather than take action on unexpected variation, he uses a technique he calls on-line control. With on-line control, economic limits (based on the loss function) are determined and any variation from target is acted upon. Key in this method is the identification of the costs of measurement and adjustment as well as the correct amount of adjustment to achieve target. Assessment: Although this technique has not been extensively documented, it is worthy of investigation. The importance of developing a good model of the loss function, however, may preclude this technique being used in the near future. The technique will work well with automated process control methods.

3. Process improvement

4 of 8

Bill Ross (303) 494-8521


Six Sigm a Associates

Taguchis methods focus not on developing the best model (as is the case with classical methods), but on getting something to work better quickly. He employs the use of fully saturated, multi-level designs run in the face of noise (while noise parameters are varied). Assessment: Perhaps Taguchi attempts to do too much, too quickly. There are

significant risks associated with his methods (i.e., high level of confounding, poor models, poor prediction possible), however his methods have had limited success. Care must be given to address these risks. In the hands of a novice these methods can have disastrous consequences. Conceptually his ideas are excellent, but some of the tools and techniques can be improved upon (using classical statistical methods).

4. Tolerancing. Taguchi believes this should be the last focus of quality efforts (after system and parameter design are completed). Changes in tolerance can be costly. Assessment: His tolerance design methodology is an excellent strategy for: a. Selecting parameters that need tolerance control. b. Identifying the appropriate levels. c. Recognizing costs associated with tight tolerances.

5 of 8

Comparison of Selected Quality Improvement Methodologies: Shainin, Taguchi and Classical

Summary Comparison The following matrix will take a simplified look at key items for selected topics.

TOPICS

SHAININ

TAGUCHI
! Process/product robustness (noise) ! Reduction of variation around target ! Efficient response variables (loss function, S/N). ! Saturated designs

CLASSICAL (Box, etal.)


! Randomization ! Sequential testing ! Best models ! Data transformation, interactions, response surface ! Fractional factorials ! Steepest ascent ! Nested designs ! Full factorial. ! Response surface. ! Multi-level. ! Central composite

! Focus on sources of variation

Philosophy

! Appropriate tolerances ! Search methods ! Rely on engineering knowledge

Screening (Sources of Variation)

! Multi-vari ! Comp./variable search

! N/A (done in conjunction with optimization). ! Saturated factorial designs ! Inner/outer arrays. ! Robust response variables

Optimization ! Full factorial

Control

! Pre-control

! On-line techniques

! Control charts

Tolerancing

! Tolerance parallelograms ! ISO plots.

! Tolerance designs (factorial).

! Stack-up. ! RMS analysis

6 of 8

Bill Ross (303) 494-8521


Six Sigm a Associates

Summary Tools There are advantages and disadvantages to any one persons methodologies. following matrix looks at advantages/disadvantages of the tools and techniques. The

TOOL

ADVANTAGES
! Identify family of variation ! Graphical ! Easy ! Identify appropriate tolerance ! Correlate customer requirement to process tolerance ! Easy ! Easy ! Statistically valid (SV).

DISADVANTAGES
! Limited in info obtained

SUGGESTED ALTERNATIVES
! Nested factorial designs

S: Multi-vari

S: Tolerance Parallelogra m

! Requires significant knowledge of product and the right X variable ! Inefficient ! Costly/time consuming ! Bad assumptions ! SV not necessarily important ! Subjectivity ! Requires knowledge of customer loss ! ! ! ! ! Inefficient Complex Questionable Confounding Complex

! Correlation/ regression analysis

S: Variable Search

! ! ! ! ! ! ! ! ! ! ! ! ! ! !

Sequential designs Fractional factorials Screening designs Nested designs Cpk. CR Cp Cpm s2

T: Loss Function

! Cost metric ! Focus on targets ! Efficient for motivating reduction in variation around target values ! Robust to noise ! Optimization in noisy environment. ! Balanced ! Efficient ! Robust to interactions ! Additivity good

T: Inner/Outer Arrays T: Orthogonal Arrays T: S/N Ratio

! Dependent on response variable ! Not always the proper transform

Blocking Randomization Covariance Standard order designs Sequential designs Coefficient of variance ! s2, x analysis ! Transformations

Note: S is Shainin; T is Taguchi.

7 of 8

Comparison of Selected Quality Improvement Methodologies: Shainin, Taguchi and Classical

Conclusion The controversy over whose method is best is spurious. Constituents of the Six

Sigma Associates believe in presenting a wide array of approaches and having the client choose those that fit their situation (or perhaps more precisely their culture). We attempt to present opposing arguments for different methodologies, but obviously may be biased because of our direct application experiences.

References Box, G.E.P. and Soren Bisgaard. The Scientific Context of Quality Improvement. Quality Progress, June 1987. Box, G.E.P, Soren Bisgaard and Conrad Fung. An Explanation and Critique of Taguchis Contributions to Quality Engineering. Quality and Reliability Engineering International, vol. 4, 1988, pp. 123-131. Gunter, Bert. Statistically Designed Experiments. Quality Progress, December 1989-August 1990. Logothetis, N., A Perspective on Shainins Approach to Experimental Design for Quality Improvement, Quality and Reliability Engineering International, vol. 6, 1990, pp. 195-202. Private Correspondence with G. Taguchi (1990), D. Shainin (1987) and G. Box (1990). Shainin, Dorian. Better Than Taguchi Orthogonal Tables. Transactions, 1986. ASQC Quality Congress

Taguchi, G. and Don Clausing. Robust Quality. Harvard Business Review, No. 90114, January-February, 1991.

8 of 8

You might also like