Several years ago, Steve Marsh, director of strategic operations at Motorola (Austin), visited Japan. Having checked into his hotel, he had only enough time to drop his baggage in his room before hurrying to dinner with a customer. He entered the elevator and pushed the button of his floor. The elevator doors closed, and, in a few seconds, opened again. Thinking that someone had pushed the elevator-request button, Steve held the door open and looked out, but no one was waiting. When he reached again to push the button for his floor, he discovered he was already there. The elevator had transported him several floors without his feeling any motion.
Being an engineer, Marsh was intrigued, so he commandeered the elevator and played awhile, moving between floors, both short and long distances, both up and down. Concentrating now, he could feel the car moving but was amazed at the lack of its jerk in both starting and stopping and at its smooth acceleration and deceleration. I don't know whether Steve made it to his dinner on time, but he later had Motorola personnel inquire about why the elevator's movement was so smooth. The explanation was simple: Fuzzy logic controls the elevators. Steve became a believer.
What is this serious technology with the funny name? In its most general definition, "fuzzy logic" is the mathematical basis, formulated and named nearly 30 years ago by University of California, Berkeley, Professor Lotfi Zadeh, that allows for shades of gray, or "fuzziness," between absolute truth and falseness. In fuzzy logic, an assertion can be partially true, and, if so, can also be partially false, to the degree that it is not true.
This simple-sounding concept has become a rather dramatic springboard. By stepping beyond the black-and-white world of bilevel logic and into the shades-of-gray world of fuzzy logic, we become better-equipped to deal with uncertainty, vagueness, imprecision, nonlinearities, and even time variance; in short, fuzzy logic helps us deal with much of the real world. Allowing logic to be fuzzy gives us greater ability to get our arms around systems that in the past had defied understanding.
Extending this definition, are fuzzy logic systems logically those that in some manner employ this mathematical basis? The answer is, "yes and no." Yes, in a general sense, and no, because, over the last few years, the phrase "fuzzy-logic system" has come almost exclusively to mean a rule-based system with rules stated as relationships between fuzzy variables. These fuzzy rule bases, which we will detail in future columns, are at the heart of many systems in diverse applications. Even in this country, which lags behind both Japan and Europe in fuzzy engineering, fuzzy products are starting to emerge. For example, General Motors Corp uses a fuzzy controller for the Saturn's automatic-transmission downshift mechanism, and Liebert Industries uses one in its LogiCool temperature and humidity controller. Metus Systems uses fuzzy logic in a fraud-detection system for health-care providers, and Delta-X Research developed and sells a Windows-based program that uses a fuzzy version of an IEEE/ANSI standard to analyze faults in large electrical-power transformers. Many more fuzzy-development projects are under way.
Having touched on what fuzzy logic is, I should also mention what it is not. It is not probability, although both fuzzy logic and probability deal with uncertainty. It is also not neural nets, although the two technologies work well together. And it is not artificial intelligence: Although adaptive fuzzy systems can and are being implemented, basic fuzzy systems are deterministic. Fuzziness does not imply learning.
Where is fuzzy logic headed? I see a historical parallel. In late 1973, I was a young engineer less than a year out of a master's program in electrical engineering and working in a group that designed satellite attitude-control systems for TRW Defense and Space Systems in Redondo Beach, CA. First-generation µPs were available, off the shelf, from Fairchild, Intel, National Semiconductor, and Rockwell International, with other vendors scrambling to introduce their devices. The trade journals, including EDN, were just starting to publish articles that gave glimpses into the µP's early potential.
I found the new technology exciting. One lunch hour, I consumed a detailed introductory article and wanted to tell someone what I had learned. In the lab, I found Bob, a senior engineer, sitting at one of the benches trouble- shooting a hybrid analog/digital breadboard with an oscilloscope.
As I recounted details from the article, Bob didn't stop probing his circuit, but he did listen. After a few minutes, I paused for a breath, and he said, "But how would we use it?" I thought for a moment and realized with some embarrassment that I didn't know. As I recall, the article had an example of a µP-controlled butcher's scale, including an illustration complete with a pork chop, but that seemed far removed from the complex control systems with which we worked.
Saying I would get back to him, I retreated to my office. Lunch hour was over, and I resumed work, still pondering Bob's question. Finally in midafternoon, it hit me. I trotted into the lab. Bob was still there, still probing. "I know what you can do with the computer on the chip," I told him. "A µP could simulate the circuit you are working on right here at the bench. You wouldn't have to separate development and analysis. It could all be done right here!" I had seen the light. It seemed so obvious. Amused, Bob looked at me and said, "I'll pass."
I don't recall how long it was before I realized that simulating the circuit was a side issue, that a µP-based system would in a few years replace the circuit Bob had designed. Bob eventually realized it too and accepted it, as well, as analog attitude-control systems became things of the past.
Moral: It is very difficult to predict what impact, if any, a new technology will have. We can remove the blinders we wear, firmly cemented in place by years of education and experience, only with great effort, even from those of us who are actively trying to do so. We can make well-thought-out, well-supported claims, but, like my µP-as-a-simulator prediction, they often miss the real mark.
We know µPs first largely replaced random logic and low-frequency analog circuits. Their real power soon became evident at the core of new, powerful user interfaces that were previously infeasible. And they are the basis for PCs and workstations—where they are often called on to perform simulations, Bob! Even a simple piece of electronic equipment now contains a µP, and complex equipment—cars, for example—have many. A decade ago that a product contained a µP and was, therefore, "computer-controlled," was still a valid selling point. Today it's ho-hum.
Fuzzy logic has the potential to have the same impact on technology over the next two decades that the µP has had over the past two. Having observed how it successfully applies to disparate applications in disparate ways, I can see fuzziness creeping into much of what we as engineers do. It will become both a design and an analysis tool in our tool bags and will positively affect many of our traditional tools, as well. In time, fuzzy logic will enable us to solve not only those problems that would be impossible to solve without it, but also those that we do not even see today.
There are those who disagree. Fuzzy opponents say that a traditional, nonfuzzy-control approach would do just as well, if not better, in controlling the Japanese elevator in which Marsh rode. They also say there is nothing fuzzy logic can do that traditional techniques cannot do as well or better. Fuzzy proponents strongly disagree. There are intelligent, well-meaning individuals on both sides. There is also rhetoric—sometimes entertaining and sometimes not—from both sides. While not wanting to appear aloof, I choose not to participate in these debates; those who use the technology—not those who debate it—will ultimately decide the issue.
I am delighted that EDN has given me the opportunity to write this column. My primary goal in doing so is to help those of you who are interested learn how to apply fuzzy logic in engineering solutions. At the advice of EDN editors, the next six or so columns will be tutorial. We will decide then which direction to take.
By "we" I also include you, the readers. I welcome your comments, observations, and suggestions.