Dear Mr. Pease:
I hope you realize what you have just done: You have secured from Dr. Genichi Taguchi the most significant statement ever made by anyone on the application of Taguchi methods: "We are not interested in any actual results..." From now on engineers should understand that this "Taguchi stuff," as you called it, is simply a game of "virtual quality" (which, like "virtual reality," is exciting and engaging as long as the goggles and gloves are on, and the player does not have to worry about what's going on in the real world).
And I would suspect thousands of those who have paid good money for Taguchi books, Taguchi courses, Taguchi workshops, Taguchi videotapes, Taguchi software packages, and what have you, would all of a sudden see the light and ask for their MONEY BACK—and that's "loss to society"!!
- T.N. Goh, Professor, National University of Singapore
Gee, that sounds like a NET GAIN to society to me.—RAP
Last week, after reading your column "What's All This Taguchi Stuff, Anyhow?" in the June 10 issue, I was in a local Computer Literacy book store and came upon a Taguchi Methods book.1 I purchased it, and have made some notes for your interest.
Perhaps of greatest interest to your readers is the book's chapter 11: "Tolerance Design and Analysis of Audio Circuits."2 The authors applied Taguchi Methods to the selection of resistors and capacitors in a third-order high-pass filter, which used one op-amp (assumed ideal), 4 resistors, and 4 capacitors.
As usual, the Taguchi Method was used in two stages: first, the "optimum" nominal values of each resistor and capacitor were determined; then, "optimum" tolerances for each component were selected. The goal was to assign component values and tolerances so that the circuit met its specifications with minimum variability.
To find the optimum nominal values of the eight components, the transfer function of the filter was computed in terms of R1-R4 and C1-C4. Three possible values of each component were chosen, giving 38 = 6561 different circuit possibilities. Then "Taguchi methods" were used to design an "experiment" whose objective was to find the best or optimum circuit among these 6561 different choices of R1-R4 and C1-C4, but by using only 27 runs, not 6561.
Two specifications were defined: the roll-off of the gain at 300 Hz versus the gain at 1 kHz, (-3 dB min) and the amount of peaking at 700 Hz versus the gain at 1 kHz (±1 dB max). Surprisingly, there was no spec for midband gain; the only two specs concern gain ratios (roll-offs).
Then the Taguchi Method was implemented. After some Latin-square orthogonal-array monkeying around, the Method calculated that the optimum component values are: R1 = 15 k, R2 = 100 k, R3 = 5.6 k, R4 = 10 k, C1 = 0.022 µF, C2 = 0.022 µF, C3 = 0.015 µF, and C4 = 0.022 µF.
Next, the assignment of tolerances was undertaken. The original tolerances on resistors were 2%, and tolerances on capacitors were 10%. The Taguchi Method concluded that capacitors C3 and C4 should be tightened up to 5% tolerance; C1 and C2 can remain at 10%, and R1-R4 can remain at 2%. Emboldened by your column, Bob, I decided to play the role of Doubting Thomas: Is it reasonable that these two specs required 2% resistors and 10% capacitors to get a working, production-worthy circuit? My intuition said, "No."
So I decided to select the components myself, using traditional (non-Taguchi) methods. My goal was to use 10% resistors and 20% capacitors, and still get 100% of all filters to meet both gain roll-off specs.
If this is possible, then the circuit could be manufactured much more cheaply, resulting in lower costs, higher sales, and in the words of Taguchi, smaller "loss to society."
I used about the same set of 38 = 6561 possibilities as the Taguchi investigators. I just chose a different final set of circuit values—R1 = 12 k, R2 = 100 k, R3 = 5.6 k, R4 = 10 k, C1 = 0.01 µF, C2 = 0.015 µF, C3 = 0.01 µF, and C4 = 0.022 µF.
I checked out my design by performing some computations. When you insert different component values into the transfer function for this circuit, it's easy to see their effect upon the response, and to see if a particular set of components passes or fails the roll-off specs.
So, using Monte Carlo techniques, with 10% nominal resistors and 20% capacitors, I generated 50,000 different high-pass filter circuits. I was pleasantly surprised to see that all 50,000 circuits "passed" the roll-off specs, using those low-cost parts.
Then I decided to check out the Taguchi-Method-derived circuit, as it was supposedly "optimized" for reduced variability. Maybe it would also yield 100% when 10% resistors and 20% capacitors were used? Here's my results: 73 of 50,000 high-pass circuits using Taguchi-optimized values failed the 300-Hz test, and 1243 failed the 700-Hz test. In other words, yield for the Taguchi-optimized circuit was 97.4%, when low-cost component tolerances (10% resistors, 20% capacitors) were used.
Bob, I am disappointed. The Taguchi Methods gave a circuit that requires higher-cost components to achieve 100% yield. Conventional (non-Taguchi) design procedures produced a circuit that had a yield of 100% when lower-cost components are used. Maybe some of your colleagues who advocate the Taguchi method can explain what went wrong?
- Taguchi Methods (Proceedings of the 1988 European Conference), Elsevier Applied Science, London & New York.
- Ibid., Chapter 11, authored by N. Logothetis and J.P. Salmon, p. 111.
-Mark Johnson, Los Altos, CA
Well, that is a good question! I always wondered if the Taguchi "optimization" methods really provided an optimum. I wonder where Mr. Logothetis got his "optimization" computer programs. Did he develop them himself, or get them from Mr. Taguchi, or from Taguchi's American Supplier Institute? Not that I care. AND this is NOT the first time that a Taguchi-expert claimed he "optimized" a circuit, when all he did was improve it a little (Phadke, Madhav, Quality Engineering using Robust Design, Prentice-Hall, Englewood Cliffs, N.J., 1979, p. 185-187). Personally, I usually use a circuit with 3 resistors and 3 capacitors to make a simple Sallen-Key filter. The response specified does not require 4 Rs and 4 Cs. I have not yet proven that I can't do it with 2 resistors and 2 capacitors. But I did check Mr. Johnson's computations, and I agree that his math is correct, and his optimum values are as good as he said, for 100% yield with cheap 20% capacitances.
NOTE: Mr. Logothetis' optimization study apparently covered and "bracketed" Mr. Johnson's "optimum," but failed to recognize it. Logothetis had the better optimum right in his sights-but failed to pull the trigger! I wonder why. I guess it just goes to show, thinking and checking your results beats trusting your computer any day.
NOTE: The Taguchi enthusiasts like to use a minimum number of computations—just 27 or 54 runs. Mr. Johnson decided to run several tens of THOUSANDS of runs. Who's living in the past?
NOTE: For a circuit that needs 5% capacitor tolerance, you have to buy good 2% capacitors so they do not drift out of a 5% spec versus time or temperature, and those capacitors really are expensive. Maybe the real "loss to society" is spending $72 on a book that says you must use tighter tolerance components when cheaper ones will actually do just fine..... —RAP
Dear Mr. Pease:
I am very interested to hear about Dr. Taguchi's comments on the "voltage regulator" that does not regulate. Let me point out a statement in a book, Introduction to Statistical Quality Control by Douglas C. Montgomery, Wiley, 1991, p. 543, regarding the specific methods of experimental design and data analysis used in the Taguchi approach to parameter design:
"On the other hand, many companies have reported success with the use of Taguchi's parameter design methods. If the methods are flawed, why do they produce successful results? Taguchi advocates often refute criticism of the methods with the remark that 'they work.'
We must remember that the 'best guess' and the 'one-factor-at-a-time' methods will also work, and occasionally they produce good results. This is no reason to claim they are good methods. Most successful applications of Taguchi's technical methods have been in industries without a history of good experimental design practice. Designers and developers were using the best-guess and one-factor-at-a-time methods (or other unstructured approaches), and since the Taguchi approach is based on the factorial design concept, it will often produce better results than the method it replaces. In other words, the factorial design is so powerful that even when it is used inefficiently, it will work better than most anything else."
So when Taguchi and his disciples wish to slough off criticism of their methods, they argue that "they work." But when Taguchi wishes to brush off criticism of his results, he says it is only the optimization method that is important. Can Taguchi have it both ways? Not after June 10, 1993. (Myself, I like the Blind Squirrel Theory—even a Blind Squirrel can find a nut occasionally. In some cases, I prefer Monte Carlo Analysis because it's easy to do and it searches at random places in the interior of Component Space, not just around the perimeter. But in other cases, factorial methods are indeed suitable, and efficient.—RAP)
Also in the same book, same page, author Montgomery observes, "A final point concerns the learning process. If the Taguchi approach to parameter design works and yields good results, we may still not know what has caused the result because of aliasing of critical interactions. In other words, we may have solved a problem (a short-term success), but we may not have gained process knowledge, which could be invaluable in future problems."
- Name Withheld On Request
I agree completely: If we trust a computer to process a huge stack of data, and it spits out a good answer (and, as we have seen, that is not a trivial matter), we have neither learned what's right nor how to go in the right direction next time. We haven't developed a feeling for the derivatives, because Mr. Taguchi told us not to bother with those—they are too difficult. He wants to keep us in the dark, and not to develop our intuition. He just wants us to trust him, and to use his method, and trust the results it puts out. Hey, maybe it's Mr. Taguchi that is the Blind Squirrel! Further, if there's any false data in the input deck, we would never realize that was what caused funny outputs. Taguchi's analysis is not likely to spot that a typo error is causing false results.—RAP.
I cringed when I read the letter (name withheld) in the May 27 Bob's Mailbox, praising the "degreed engineer." I'll give you my experience for what it's worth.
I run a small electronics engineering company in the North of England. Two of us design our products, both software and hardware. We both learned by "reading hobby magazines" and generally devoting ourselves to the subject at hand.
Actually, we've been reading the magazines since we were kids and I'm now approaching 40. I know very little about higher mathematics and virtually nothing about laminar flow. But our speech products have won awards from the chip manufacturers, and our single-chip designs are in use in demanding applications the world over.
Ten years ago, we developed a videotape control system so sophisticated that the press found it necessary to call it "an amalgamation of Japanese and American technology," though it was developed from scratch in the one-time mining area of Newcastle Upon Tyne in England.
Since then, we have gone from product to product with no help from universities or outside bodies. Our latest product has received a glowing review from Jerry Pournelle of BYTE magazine, as well as virtually every U.K. computing magazine.
I have run local computer clubs in my spare time and I know technical folk worldwide. I'm not showing off, but am qualifying my viewpoint: I firmly believe that when it comes to producing the goods, the only difference between a "highly qualified" expert and a self-taught expert is a piece of paper. What is important is dedication, clear thinking, and more dedication. You will often find lots of that in duly-qualified engineers, but only because it was there in the first place.
I certainly wouldn't want to put anyone off going for qualifications. By all means, it will help your salary and enrich your knowledge. But don't think for a minute that a piece of paper is going to turn an ordinary guy into a star designer... absolutely not! Never has, never will. Name NOT withheld!
- Peter Scargill, Newcastle Upon Tyne, England
Normally I shriek and holler in favor of clear thinking, but I'll admit that dedication is an essential ingredient! —RAP
The idea that a degree somehow makes a person a better worker is absurd! Also, the fact that someone has a degree from an unaccredited or unrecognized school does not mean that person is a bad worker.
For part of my two years as an Army draftee, I worked as an "electrical engineer assistant." My degree was in physics, with only one class in electronics. I worked side-by-side with several EEs, including one from MIT. There was one other draftee whose degree was in physics.
Both of us physics graduates had learned electronics by hands-on personal experiments. The EEs learned all they knew in the classroom and limited lab work. Yet it turned out we physicists could run circles around all of the others.
The need for a degree is, in my opinion, very overrated. When will industry and government recognize that pay should be related to performance? There should not be any artificial pay caps based on education. If the person has the knowledge and experience to get the job done correctly and in a timely manner, that person should be paid the same as another without reference to education.
At this time, I have 38 years of experience, which has qualified me in ways no classroom at any level can do. But I am (obviously) over 40 and therefore "outdated." People forget that we "old-timers" have not only all that work experience, but also have read—yes, studied—thousands of pages of technical material and books to keep up.
Yet a new hire today that has a PhD can be hired at a higher grade than I am, despite the fact that many new PhDs don't have the foggiest idea what goes on in the real world of work. Performance is NOT the name of the game in many places. It is too often who you know and not what you know how to do that gets the cake.
Now don't get me wrong. Education serves a very useful place. But on-the-job education is also very useful. The point is that automatically separating engineers and technicians based solely on "formal" education, and ignoring the real-world, on-the-job education, is the cause of too much arbitrary discrimination.
- Gordon Simkin, La Luz, New Mexico
I'll pay for a can-do guy everytime, but the world does generally pay a premium for educational status, all things being equal. I can't change the world.—RAP
In the May 27 issue, your column had a letter from an individual who complained about the whining of nondegreed engineers. Your answer proved what I have often suspected: You truly are a sanctimonious cretin!
You stated, "I don't find all that much correlation between the degree and the abilities. Surely a lot of degreed engineers can't remember how to compute the rms value of a lumpy waveform." I can't believe this! Just how many non-degreed engineers do you figure can compute that? If you do a little checking, you'll find that almost without exception those who claim that degrees are of dubious value are those who don't have one (but wish they did). Similarly, I have observed that those with only bachelor's degrees often feel graduate degrees are not needed. (Gee, Bob, see any correlation there??)
Engineers with graduate degrees have learned design and analysis techniques that those without degrees (or those only with bachelor's degrees) have never even heard of. How many non-degreed engineers have a sufficient background to design digital filters with attenuation-band adjustments> Or to design multifunction pipelines, lead-lag compensators, etc.?
...The purpose of this letter is not to vilify non-degreed engineers; it is simply to put things into perspective. There are gainful employment opportunities in the engineering field for people with little formal education and for those with PhDs. But, this doesn't mean all should be accorded the same status or prestige. If non-degreed engineers want to be treated the same as degreed engineers, then they should stop whining, get off their behinds, and go to school.
- Name Withheld (Disgusted)
I'm sorry, but I don't agree with you at all—not on anything. Almost every "non-degreed engineers" knows five or ten amazing, useful things that you or I couldn't hack. Good for them! Everybody knows something! Aren't you glad I'm not your boss? I'm glad you aren't my boss.—RAP