Dear Mr. Pease:
I am one of the young promoters of Fuzzy Logic, or "F.L. Guys" you spoke about in your "What's All This Fuzzy Logic Stuff, Anyhow? (Part II and III)." But you may find it somewhat surprising that I agree with much of what you said. There are far too many people out there making exaggerated claims about how F.L. can be applied and what benefits should be expected. It is truly unfortunate that the hyping of F.L. as a panacea has obscured the fact that, properly applied, F.L. can provide real benefits.
Many of the people who have given F.L. a bad name are the academic types who, as you stated, speak as if they are "preaching to the converted." I am as tired of hearing about how F.L. will give you "more robust control" as you are about hearing about probability. Working for a factory automation company, I do not have the luxury of strictly dealing with theory (like the academics); I have to provide tangible benefits.
Omron has been a leader in the development of F.L. in various forms - microcontrollers, software and development tools, controllers, and turnkey systems. In the U.S., our focus has been on automation products. We are not actively marketing chips and software at this time.
It's important to note that in the introductory stages of anything new, it's easy to get carried away with the potential impact. (America's space program serves as an example. Weren't we supposed to have colonized Mars by now?) The promise of F.L. was no different. Today, though, as suppliers and supporters become more familiar with F.L., we've also become more aware of its practical applications and implications. Much of the boasting has ceased, though not totally disappeared.
At Omron, we've never made a blanket statement that F.L. is superior to conventional systems. The fact that our single-loop controllers (the ready-to-use 1/4 and 1/8 DIN types, not microcontrollers) use a hybrid of PID and F.L. processing speaks to this point. PID control alone is a perfectly acceptable control method in about 80% of the process applications we see. However, the other 20% are critical applications that are susceptible to process upsets, where the PID parameters cannot be tuned to provide the desired combination of responsiveness and stability. With our controllers, the PID constants can be tuned for stability, while the F.L. part of the algorithm deals with responding to the disturbance.
Our customers using this particular controller report significant improvement of process control using the F.L.-based product over the PID product (also Omron's). These customers are in the food, pulp and paper, pharmaceutical, medical products, semiconductor, and other industries. When one of these customers allows us to write about the application, we'll be sure to send you the details. Until then, we accept customer orders for more F.L.-based controllers as proof that they work better for them in their applications.
Omron's tests of performance in controlled environments yielded a 50% improvement in controller response time over PID-only controllers. We fully understand, though, that this improvement in response time is important for only some of our customers. But, for those customers, the only really important thing is that the F.L. product solved their problem.
Like you, Omron also disagrees with the idea that F.L. users don't need models to develop a system. Our seminars on F.L. emphasize the need for a thorough understanding of the system to be controlled, and a clear understanding of the overall objectives of the project before F.L. can be successfully applied. Our stance is if a conventional controller can provide the necessary performance, USE IT! If not, a F.L.-based product may solve the problem.
As with virtually everything else in life, it comes down to a cost-benefit compromise. You mentioned in your column that a PID algorithm could be written to accomplish the same performance as the F.L. solution; all that's needed is a faster computer. The question then is how fast? And is this faster computer available at a cost and engineering development level that's acceptable to the customer?
Thanks for helping to bring the benefits of Fuzzy Logic into focus. It is not a panacea as some zealots might claim, nor is it a boondoggle of unfulfilled promises. Fuzzy Logic is simply another tool in the engineer's toolbox - an alternate control processing technique for use alone or in combination with other techniques for applications that have resisted other methods, or that require F.L.'s innate ability to capture the intuition and judgment of a skilled expert.
ROBERT NEAGLE, Product Marketing Specialist, Omron Electronics Inc., Schaumberg, Ill.
Thank you for the Sanity Check. When you and I are able to provide more details of where F.L. can provide real benefits, we'll speak up. However, some experts tell me that F.L. needs less computing power to do a job (vs. conventional controllers). And other experts say that in many cases, F.L. needs more computing power. Nothing simple in the whole world! - RAP
Dear Mr. Pease:
While reading your recent "What's All This Fuzzy Logic Stuff, Anyhow? (Part II)" column, I was reminded of a past experience that may pertain somewhat to the state of the Fuzzy Logic community. Some years ago, at a different firm, we hired a software engineer, Bob. In a strange sort of way, Bob was very bright. He was to a great extent self-educated. While that is not a bad thing, Bob's knowledge ended up having unexpected gaps, which, unfortunately, he had little interest in filling. Being clever, he had developed many interesting techniques to deal with design problems over the years. He was very proud of these, and considered them to be original and unique. But, unknown to Bob, many of them were simply traditional designs that Bob had rediscovered.
I remember one case in which he proudly described how he would make his software "anticipate" the arrival of a certain external event that occurred at fairly uniform, although not exactly known, time intervals. He began a long discussion explaining how this miracle worked, using a combination of the system's local timebase and some of his software. Part way through, I interrupted him, and tried to tell him that the thing he was describing had a name. It was called a phase-locked loop. Bob countered that I couldn't have understood him correctly. He had invented this thing two years ago, and he had not named it a PLL.
I explained that I understood correctly, and although his terms were unconventional, he was indeed describing a PLL. PLLs were not invented two years ago, but have been used for about half a century. I explained that although the original ones were analog circuits, purely digital ones also existed. Using a fixed clock and a variable modulus counter (rather than an analog VCO), they were very common in systems that recovered synchronous serial data, and could be found in various popular datacom peripheral chips.
I explained that Bob's code was simply a software implementation of one of these digital PLLs. These software implementations were not unusual, being found in some automobile-engine timing-control software, among other places. Of course, Bob had never seen anyone else's digital PLL, although he had used products full of them. I suspect he may even have seen an analog one, but never made the connection between that circuit and his "invention."
Bob was very resistant to my claim that textbooks contained well-documented analysis techniques for his invention, since he, the inventor, had to tune his creation by trial and error. But I don't think he ever had the slightest interest in looking through such texts. He didn't accept that we already understood his invention in detail, or that we already had enough experience with PLLs to know that it was not the best answer to the current problem.
Bob deserves a lot of credit for having independently reinvented something as wonderful as a PLL. But, because of his limited background, he failed to understand that what he had created was not all that impressive, and was very unreceptive to having this pointed out to him.
I suspect that some F.L. proponents may have similarly weak backgrounds, and may not have seen examples of well-designed traditional control systems. Bob's PLL implementation was not immediately recognizable as a PLL, even though in the end it proved to be a classical device. I further suspect that some of the things we see in Fuzzy systems are, in fact, techniques that have been in use for some time to enhance traditional controllers, but mapped into a different domain where they are no longer recognizable.
I don't mean to imply that F.L. has not embodied classical control equations. I am referring to the clamps, limiters, nonlinear signal conditioners, "anti-windup" circuits, hysteresis, feedforward, etc., which are often part of real (if not textbook) controllers. Some of these features long predate software, being found in analog and even some ancient pneumatic/mechanical controllers.
An advantage of F.L. may be that it can produce a very good controller by following some analytical design procedure. However, the enhancements added to traditional controllers to improve their performance were generally the result of the skill and experience of the engineer.
There have always been physical systems that traditional control textbooks regard as more or less uncontrollable, or at least very difficult to control. When such systems are easily brought under control by F.L. techniques (and some have been), we have an accomplishment worthy of respect. But, most of the commercial products claiming to use F.L. today do not seem to fall into this classification. Therefore, I find myself in agreement that F.L. is not quite as revolutionary as its proponents claim.
ERIC KINAST, Engineering Manager, Sonntek Inc., Woodcliff Lake, N.J.
Every engineer must be prepared to invent a circuit or function that he/she has never seen before. If it then turns out, it's already been done before, there's no point in getting mad! As I mentioned in my 1992 columns (ELECTRONIC DESIGN, Jan. 9, p.41 and Nov. 12, p.80) the PLL was developed as early as 1783, for adjusting clocks and watches. Fuzzy Logic may be a useful tool to help us handle nonlinearity. But it doesn't usually give ideal results just by pushing a button - the engineer still has to provide skill, experience, and thinking. - RAP
In your femtoamp column in the Sept. 2 issue, you mentioned placing guard rings on the bottom and possibly the top of the pc board to stop stray leakage currents from reaching sensitive nodes. While working at a company here in Cleveland that makes femtoamp meters, I also learned to "stitch" the pc board. In other words, to put numerous plated through-holes between the top and bottom guard rings. We found conduction down some of the fibers under the surface, so we added the holes to break these conduction paths. I don't know if this was just a process problem making pc boards 15 years ago, or if it still exists. But I still stitch boards around sensitive nodes.
GARY BERGSTROM, Bergstrom Consulting, Chagrin Falls, Ohio
I've stitched my pc boards for different reasons, but you are right in this case. Thanks. - RAP
All for now. / Comments invited!
RAP / Robert A. Pease / Engineer
Address: Mail Stop D2597A, National Semiconductor, P.O. Box 58090, Santa Clara, CA 95052-8090