Electronic Design

Bob's Mailbox

No other Pease Porridge column has generated more reader response than "What's All This Taguchi Stuff, Anyhow?" which appeared in the June 25 issue.  Bob's hot on the trail of those elusive Widget amplifiers, and will be devoting another column to the Taguchi method in the near future.  In the meantime, here's a sampling of our readers' opinions on the subject.

Dear Mr. Pease:

Your article "What's All This Taguchi Stuff, Anyhow?" is refreshing reading.  I suppose one could look at the Taguchi issue better if distinctions are made among:  (a) Taguchi the Master, (b) Taguchi Methods, (c) Taguchi Preachers, and (d) Taguchi Disciples.  Then it would be clearer which are the ones that ought to get out of real labs that deal with real systems, and be asked not to sell any more magical data and beautiful parables concocted supposedly for the salvation of engineers.

I am doing research on this Taguchi phenomenon, and would like to keep in touch with you and everyone else that has experience with the "Taguchi Methods."

T.N. Goh, Professor, National University Of Singapore

I, too, am concerned about those magical data and miracle cures and I'll share my ideas and letters with you.  More later.-RAP

Dear Mr. Pease

Wrap it up!  Your two-page put down of Taguchi only puts me off my Porridge.

The Taguchi approach explores the whole design space and enables many factors to be evaluated for the significance of their contribution.  Confirmatory test work is always recommended.

Please recheck your own signal-to-noise ratio.

Richard W. Gilbert, Product Engineering Manager, Basingstoke, England

I've been checking my work.  My S/N ratio is fine.  It just bothers me when people think they can use Taguchi's methods on nongaussian data, just because they see Dr. Taguchi doing it.-RAP

Dear Bob:

I just read your article on the "Taguchi Stuff."  It is very well written and hits the nail directly on the head.  I thought I was the only one on the planet that was suspect of this whole thing.

As a senior process engineer for a company that does injection molding, occasionally I'm involved with troubleshooting extremely difficult parts to injection mold.  The more time I spent in our process development lab, the more I saw of this "Taguchi Stuff."  It had become a monster from the simple little method that it was originally intended for (I personally think it would be good for making pancakes if you had no recipe, just the Bisquick, eggs, and some milk).

On a serious note, the type of molding machinery we have is some of the most sophisticated in the world and has many closed-loop feedback systems that deal with outside disturbances, such as material viscosity.  I found out that our process technicians were trying to do these "Taguchi Tests" that were designed by quality engineers.  In order to run these tests, though, they had to "fool" some of the closed-loop systems.  In short, they generated information that could never be used in manufacturing or the test wouldn't even run.

I asked a lot of technicians and process engineers around the company if they ever ran a Taguchi test that lead to a better process than the one they had.  The answer was "Yea, once" but only because one of the tests itself was a better process (I'd call that luck).

The tests still go on, but not on any of my projects.  I believe strongly that common sense, intuition, logic, reason, and experience bring much more to the party than this "Taguchi Stuff" will ever bring.

Edward J. Devault, Sr. Process Engineer, Nypro Inc., Clinton, Mass.

Obviously we agree on most aspects.  There are many good ways to optimize a system.  But the Taguchi method does not provide me any insights.  Thanks for your real-world observations.-RAP

Dear Mr. Pease

I was sorry to read about your negative feelings concerning the Taguchi Method.  I was, however, pleased to note your scientific curiosity as well as your desire to learn a new method.

I have been working in the field of the Taguchi method, also known as the Robust Design method, for the last twelve years and I have seen it work in a wide range of industries: telecommunications, microelectronics, xerography (electrophotography), aerospace subsystems, defense electronics, steel making, software testing, and software development.  These applications range from early stages of technology development, product design, and manufacturing process design, to fire fighting in manufacturing operations.  Of course, it goes without saying that Robust Design is not a tool that solves all problems.  Selecting an appropriate tool and applying it correctly are essential for getting good results.  When applied correctly in appropriate situations, it has yielded excellent benefits.  The same thing can be said about statistical methods.  They, too, have yielded excellent benefits.

I would like to suggest a few references for you.  Keeping modesty aside, I would like to refer you to the book Quality Engineering Using Robust Design, by M.S. Phadke, Prentice-Hall, 1989, for a clear explanation of the Robust Design method.  It also has some case studies relevant to the microelectronics field.  The book Quality by Design:  Taguchi Methods and U.S. Industry, by Lance Ealey, ASI Press, 1988, describes the experiences of several industries in the United States and other countries in applying the Robust Design method.

Dr. Madhav S. Phadke, President, Phadke Associates, Inc., Tinton Falls, N.  J.

I am definitely interested in learning about other people's ideas and books for "robust design."  Engineers have been designing mechanical and electronic products for many years.  Sometimes good "robust" results were achieved.  How did we ever do this before Dr. Taguchi came along?-RAP

Dr. Mr. Pease,

Regarding Taguchi, I couldn't agree more and I'd like to relate a similar experience.  My company is putting all of its employees through "Five Lambda" training as part of its quest for a well-known quality award.

In this class, we were shown a video tape in which a PhD type from another company (from which we purchased the course) showed us how to use Five Lambda techniques to optimize the design of a stopband filter.

He chose one leg of the filter.  First he said his technique would eliminate the archaic "tweaking" capacitor (he never followed through with this).  I contend that most filters have to be tweaked because of unpredictable effects, like mutual coupling between components, putting the lid on the can, etc.  Second, he went to great lengths to show that the value of the resistor had little effect on what he called the "Robustness" of the circuit.  The idiot didn't even realize that there was no actual resistor in the circuit, just the resistance of the coil wire.  Third, he optimized (by statistical methods) the value of L and C for best input impedance and passband insertion loss.  These are not the parameters that define the "Goodness" of such a filter.

I pointed all of this out to the instructor and suggested that this man and his company's very expensive course have no credibility with me.

Name Withheld, Silicon Valley, Calif.

Some of these "experts" take themselves terribly seriously.  They claim to be experts in EVERYTHING, even topics they really do not understand.-RAP

Dear Mr. Pease:

In regard to "What's All This Taguchi Stuff, Anyhow?," your article really misses the point.  You have clearly demonstrated a total lack of understanding of the process.  As a case point, the Taguchi method is just one tool in a toolbox of tricks, but you treat it as though it totally replaces everything in the lab, including common sense.

No, Mr. Pease, that won't work in my lab.  Take that Pease Stuff and shove it in the round file.  I don't need that Pease dividend or any others like it.

On a more positive note, I'd like to hear more about the "Pease Stuff." If you don't use the Taguchi method as one of the tools in your arsenal, what do you use?  How do you perform process optimization, design for higher yields, and process control?

Charles Jackson, TRW, Redondo Beach, Calif.

Ah, but peddlers of the Taguchi method don't say it is just one tool in the toolbox.  They claim it's the only tool you'll ever need.  Myself, I optimize my work using 2D scatter plots.  I use my eyes, skeptically (not any mindless computer) to look for correlation, trends, troubles, and solutions.-RAP

Dear Mr. Pease:

I hope that all readers of your column and those who have attended similar Taguchi lectures are equally as skeptical.  I have attended several such lectures and have come away from each with a great deal of frustration because the Taguchi method seems to oversimplify things and seems so quick to ignore and discount things that would 'muck up' the application of the method.

Similarly, the examples used are always nonrealistic scenarios (such as the magic widget amplifier); and the solution could be easily figured out without the Taguchi Method.

One other short coming that you may wish to point out to your readers is that the Taguchi Method (as it has been explained to me) completely discounts the use of random samples and tests of processes and their outputs because it is assumed that the process is always under control-therefore nothing random can happen.  So!  Our processes are therefore completely under control-we have just chosen lousy yield tolerances?  We need the Taguchi Method to tell us to tweak things to get tighter SPC?  I don't think so!

Finally, there is a ray of hope for those who like to use a methodology in their quest for better quality control:  "Design of Experiments," by Keki R. Bhote, is a method I have found to be much more logical and scientific as well as simple to learn and apply.

Steven J. Vornsand, Audio Engineer, Dukane Corp., Saint Charles, III.

Isn't it funny how these gurus set up a simple problem, to demonstrate the superiority of their methods.  And when the methods, after massive amounts of work, give useless or misleading answers, then what?-RAP

Dear Mr. Pease:

You are clearly very frustrated with the poor presentation you got on the Taguchi method.  You also mentioned the ridiculous attitude of the professor you contacted.  Your criticism mostly centered around the inappropriate examples cited during the lessons and the frustrated interactions you have with the professor.  Lastly, you do not understand the techniques and you do not allow your people to practice the techniques in your area.

However, from a scientific view point, that criticism doesn't seem applicable to the Taguchi way of experiment design per se.  Moreover, I cannot agree with you on the following arguments:

(1) You wrote:  "The method implies you can check the design-center zero offset in the morning, and design-center full-scale of the meter at noon, and take data all day long.  And you'll never have to recheck the design-center zero reading or the gain, because that would be wasteful and redundant."  And you wrote that your people always include a sanity check.

To my understanding, the Method does not mean what you said.  The "sanity check" can be included by simply putting it in as an extra factor in the Noise Matrix.  In this way, you can ensure the corresponding repeatability in your data, and at the same time not expand the number of experiments.

The Method is just providing us with some techniques which may or may not be useful, depending on the particular situation.  If people do not understand them, and use them wrongly, obviously they will generate wrong data from the experiment.  Similarly, if people are using calculus wrongly, and it generates wrong results, this only means there's a problem with the way they are applying it.  It's nothing to do with the calculus itself.

(2) In another paragraph you wrote: "...they would manipulate the variables they could control-the temperature of the plastic, the pressure, the molding time.  But they would not worry about the viscosity of the incoming plastic material because they didn't have any control of this.  So they would ignore it."

I think this is seriously violating the spirit of conducting experiments and as far as I know, is not found in Taguchi's literature.  The Method provides a new way of designing experiments, but it's not a substitute for the basic skills of conducting experiments.  This might be a mistake or misquote from the professor you mentioned.

I cannot see why you developed the hostility towards the use of statistical methods, although I agree with you that there will be problems if designs of experiments are "dictated" by statisticians, or scientists, or engineers...  This is not the time to talk about discrimination against a particular profession or race.  The world functions well through the cooperation of different professions and people instead of unhealthy hostility.

I think a new method should be judged on a merit basis.  If it does not violate scientific principles, then it is a sound method to use, whether it also concentrates on reducing "manufacturing costs" or "loss to society."

To what I see, the Taguchi method is one way of doing DOE (design of experiments).  It is a sound and practical method.  It also gives us efficiency by adopting different techniques from various fields, such as orthogonal matrices, graph theory, ANOVA, and so on.

The fact that you do not understand the technique does not mean that the method is no good.  This only means that you have to work harder.  Maybe consulting a better source will give you more enlightenment.

Wallace Yong, Advanced Engineering, Seagate Technology International, Singapore

Ah, but we refuse to use it because we understand it all too well.  And soon, enlightenment will be provided for all.-RAP

 

Hide comments

Comments

  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Publish