These books are published

by

Taylor & Francis - CRC Press

London • New York • Oslo • Philadelphia • Singapore • Stockholm
UK Head Office: 2 Park Square, Milton Park, Abingdon, Oxford OX14 4RN

The Realities of Taguchi

by Thomas B. Barker

There has been a considerable amount of misunderstanding regarding the philosophies, methods, and techniques attributed to the Japanese engineer, Genichi Taguchi. When I first heard about Taguchi over a dozen years ago while working at Xerox, I was skeptical and rather cynical of his use of three-level designs. After all, the central composite design was developed to combat the failings of the three-level’s non-rotatability and high degree of confounding. Concerned that Xerox was being sold a “pig in a poke,” I called Stu Hunter and asked him if he had heard anything about this “Taguchi” stuff. He replied in the affirmative and went further to say that he was spending a considerable amount of time doing “damage control” in the wake of Taguchi’s spreading of misinformation about experimental design and analysis. I also called an associate at Kodak and inquired and learned that he had not yet heard of Taguchi. Two weeks later, in the late fall of 1982 I was given the assignment of “figuring out what this Taguchi Method® was all about.” I am still working on that assignment and in the spirit of continuous improvement, I expect to extend my efforts right into the next millennium.

The Fantasy of Taguchi According to Lance Ealey, a professional writer and publicist,Taguchi

invented a simplified approach to experimental design and these methods are the only way to improve the quality of manufactured goods. Ideas like these immediately inflame the scientific and statistical community. In my review (1) of Ealey’s book on Taguchi, I labeled many of the statements as propaganda. So what are the realities?

-Reality #1: Despite the fact that “Taguchi Methods®” is a registered trademark of the American Supplier Institute, Dr. Taguchi insists that we call his approach “Quality Engineering by Design” (QED). He does not like the personification of his work and emphasizes the fact the he did not invent experimental design. He gives full credit to Fisher and others for the development of experimental design methods. This leads to:

-Reality #2: Taguchi does not mean experimental design. So many of us now use the word “Taguchi” as a synonym for designed experimentation. Experimental design is but a small part of the system Dr. Taguchi has developed over the past 55 years to support quality engineering efforts. His main effort now is working on fundamental engineering measures like energy. He believes that with the conservation of energy, we will have higher quality and more reliable manufactured products.

Of course, the major attraction of the QED methodology to management was the conservation of money. Dr. TaguchiÕs application of the loss function to engineering problems provided the link between management and engineering that had been missing for so long. Now if we have a loss function, we can measure in monetary units (a universal language) the penalty for variation and the penalty for being off the target. But the loss function is often not easily available. This leads to:

-Reality #3: The signal to noise ratio (S/N) is a utility function that is merely a transform of the loss function. Many statisticians have the mistaken idea that the S/N is a transform of the type used to allow analysis when the data does not fit the ordinary “assumptions” such as normality, homogeneity, and independence of the error variance. In Chapter 2 of my book (2) on the Taguchi Approach, a clear link is established between the loss function and the S/N. The S/N does not require the derivation of a loss function to find the factor settings that will minimize variation and simultaneously put the product characteristic on the target. When the S/N is

maximized, we have minimized the loss and havefound the optimal quality that the system can deliver. This may sound like magic and it leads to:

-Reality #4: The QED approach is not magic! Like any good engineering effort, QED takes dedicated work and some prior knowledge of the process that is being investigated. Many engineers who have completed a short 1 to 2 day overview seminar on the QED process try the methods and often fail to produce the promised results. They grab the smallest experimental design configuration (like an 8 run 2^7-4 that is called the L8) and run with it believing they are efficient because they looked at so many factors with so few runs. They never learned that efficiency in experimentation is a double edged sword that needs to be honed and balanced between obtaining the required information and doing so at the least expenditure of resources. In neglecting the required information edge of the experimental sword, they miss the opportunity to make continuous improvements and in failure, discard the entire QED method as another passing fad. This is a tragic occurrence that has set back engineering progress and leads to:

-Reality #6: Taguchi’s orthogonal arrays are actually representations of ordinary fractional factorial designs. The only problem with his presentation of these design configurations, is the absence of the information on the complete confounding patterns in these designs. His orthogonal arrays provide the necessary information to prevent stupid applications, but they do not provide sufficient information to allow the wise applications we need for complex systems that include interactions. This leads us to the final reality.

-Reality #7: The QED robustification process based on Taguchi’s philosophy of System Design, Parameter Design, and Tolerance Design is completely operational using the set of experimental design tools developed by statisticians in the Western World. So to deny using Taguchi’s philosophy just because you do not wish to use his incomplete tools of experimentation, would be like throwing the baby out with the bath water. The ultimate reality is that we can use the robust QED methods with ordinary classical experimental design configurations.

References

1. Quality Progress, Book Reviews, Feb., 1990, p109.

2. Barker, T.B. (1990). Engineering Quality by Design - Interpreting the Taguchi Approach, Marcel Dekker, Inc. New York.