« iepriekšējāTurpināt »
computations to form the rules actually compensates for the turbulence in the television, uses fuzzy logic for image appearing in his controller. The reason liquid, as well as the totally erratic stabilization. This happens to be the for this approach is that the composi- movements of the mouse. Thus in the very first video camera to appear tional rule involves a matrix operation latter, a claim could be made for exe- with image stabilization capability. that cannot normally be performed fast cuting control even beyond nonlinear- Matsushita, Hitachi, Sanyo, and Sharp enough on a standard digital computer. ity and into truly random or "chaotic" now have their own “fuzzy washing In Yamakawa's system, this problem domains.
machine,” which automatically adjusts was overcome by designing a chip spe- Before reporting these results, the washing cycle in response to size of cifically for this computation. An Yamakawa applied for patents on his load, type of dirt (soil versus grease), important feature of Yamakawa's chips in Japan, the United States, and amount of dirt, and type of fabric. In approach, moreover, is the use of analog several European nations. He then Matsushita's machine, the type and techniques, rather than digital. This proceeded to trade his patents to sev- amount of dirt are detected by means was done because the elementary opera- eral Japanese corporations in return of light sensors, which also use fuzzy tions employed in the compositional for their subsidizing a laboratory in controls. Other products using fuzzy rule of inference, and for the most part which he could continue his research. control include vacuum cleaners, air also in the defuzzification operation, (Japanese university professors are conditioners, electric fans, and hot plates. are the arithmetic max and min, which not allowed to make money outside of One senses that the possibility for such can be implemented so as to run much their duties as a faculty member.) applications is virtually endless. faster on an analog device.
Omron, a major producer of second Another, now famous, application is a The controller presented at IFSA-87 tier electronic devices, was a major road tunnel ventilation system, also used seven rule chips and one defuzzi- proponent and has subsequently decided designed by Yasunobu at Hitachi. fier chip, and it demonstrated balanc- to invest heavily in fuzzy control Omron A somewhat more ambitious projing response speeds approaching 100 has been rapidly expanding on
ect is the voice-controlled helicopter times faster than those heretofore Yamakawa's original designs, produc- being developed by Michio Sugeno at accomplished by a conventional PID inga host of new chips, both analog and the Tokyo Institute of Technology. Here controller. This result generated a flurry digital, and churning out scores of the objective is to develop a helicopter of commentary, including a few nega- applications. Due to its purchase of that responds to voice commands like tive responses both from without and Yamakawa's patents, in fact, it has “hover,” “forward,” “back," "left," within the fuzzy systems community. recently become the first Japanese “right,” “up," and "down,” where each The latter stemmed from the fact that corporation to ever obtain a U.S. patent.
corporation to ever obtain a U.S. patent such operation is handled automatithe controller only maintained verti- As of July 1991, Omron boasted 700 cally via fuzzy logic. Sugeno has succal, and not horizontal, stability of the patents for fuzzy logic devices either cessfully accomplished all functions inverted pendulum, whereas the classi- acquired, pending, or in application with a 1-meter model and is now working cal problem entails both. Moreover, it Most of these devices either have on a 3-meter model. As of August, he was shown rather easily that, with that appeared, or will appear, in commer- had achieved hovering and was confident particular system, accomplishing both cial products. Three or four dozen alone that the other operations could be was impossible. Hence Yamakawa are earmarked for use in automobiles, accomplished as well. Hovering is well suffered criticism for publishing results e.g., antilock brakes, automatic trans- known to be a very difficult stability that were as yet incomplete.
missions, impact warning and moni- problem; beginning helicopter pilots Less than a year later, however, toring, windshield washers, and light typically train for weeks before being Yamakawa was able to vindicate him- dimmers. Omron is also incorporating able to do this manually. Hence, sucself by producing a system with only fuzzy control into products for use in cessfully automating the hovering four additional rule chips that performed industrial and manufacturing processes. operation is in itself a very impressive both vertical and horizontal stabiliza- Numerous commercial products result. (See the article by D.K. Kahaner tion at the same speed as before. Since using fuzzy technology are currently and D.G. Schwartz, “Fuzzy Helicopter that time, Yamakawa has demonstrated available in Japan, and a few are now Flight Control," Scientific Information the robustness of his system for non- being marketed in the United States Bulletin 16(4), 13-15 (1991).] linear control by attaching a small plat- and Europe. Canon uses a fuzzy con- These few examples illustrate the form to the top of the inverted pendu
troller in the autofocus mechanism of variety of possible applications for fuzzy lum, on which is then placed a wine its new 8-mm movie camera. The logic control. Japanese manufacturers glass filled with liquid, or even a live Matsushita/Panasonic “Palmcorder," are in fact now opting for fuzzy conwhite mouse. The controller nicely currently being promoted on U.S. trollers even where conventional controllers would serve just as well. problems. Therefore, from a purely and the vice president is Hideo Tanaka The reasons are that simple fuzzy logic pragmatic standpoint, it makes sense of Osaka Prefecture University. controllers are much easier to design, to focus on reaping the commercial The society has a few regional require fewer electronic components, benefits of what has already been done chapters, and on 18 May I attended the and are therefore cheaper to produce. and to leave for the future the more
and to leave for the future the more meeting of the Kansai Branch in Osaka. The problem of how to design more challenging theoretical issues.
There I noted from the talks given that complex controllers, however, has only The possibilities for future work, the fuzzy systems group in Japan is not recently met with what appears to be a leading to far more sophisticated completely focused within that area practical solution. Typically the most logic-based controllers, are nonetheless but also overlaps with the broader realms difficult part of designing any fuzzy very clear. This will amount to moving of AI. Two talks were given, one on logic controller lies in selecting the from simple one-step rule-based systems neural nets by Kazuyoshi Tsutumi of fuzzy sets to use for the meanings of the to systems employing multi-step rea- Ryukoku University and one on caselinguistic terms appearing in the infer- soning--i.e., rule chaining, together based reasoning by Tetsuo Sawaragi of ence rules. As the number of rules grows with the necessary truth maintenance Kyoto University. large, the trial-and-error method of systems--which are integrated with other During 12-14 June 1991, I attended selecting the optimal collection of knowledge representation, reasoning, and gave a presentation at the seventh membership functions becomes less and learning schemes (e.g., semantic annual SOFT symposium, held in feasible. Somewhat of a breakthrough nets, frames, conceptual graphs, neural Nagoya. Total attendance was 500, of on this problem appears to have been nets, and case-based reasoning). Tak- which approximately 100 were from achieved by Akira Maeda at Hitachi's ing the theory to this next stage will corporations. There were 165 papers System Development Laboratory. accordingly require progress in a num presented, covering the entire specMaeda's idea is to use a form of neural ber of important subareas before real- trum of fuzzy systems research, but with net with back propagation to learn the izing the more advanced levels of auto- a lessor emphasis on theory than on needed membership functions from a matic control.
applications. Most Japanese researchers set of training examples. As a test case,
involved in fuzzy systems research are Maeda and his coworkers applied this OTHER FUZZY
engineers, so that most presentations technique to the development of a SYSTEMS RESEARCH
were in the engineering disciplines. controller that had been designed pre
Recognition of the importance of theviously by trial-and-error. Using this As may be seen from the attached oretical work was evident, however, by technique, they were able to accom- bibliography, current fuzzy systems the invitation of Satoko Titani (Chuba plish in 1 month what had formerly research encompasses a large variety University) to give a plenary talk about taken 6 months.
of topics, far too numerous to be her work with Gaishi Takeuchi on forA common opinion among Japanese covered thoroughly in this short report. malizations of fuzzy logic. researchers is that most of the impor- Therefore, the following sections will A plenary talk by Masao Mukaidono tant theoretical work in fuzzy control focus rather on the organizations, titled “Fuzzy' and Al” revealed that has now been completed and that the research institutes, universities, and the Japanese AI community has folnext step is up to the commercial corporate laboratories with which I
corporate laboratories with which I lowed the lead of its American counmanufacturers, i.e., to start churning became familiar during my visit.
became familiar during my visit. terpart in adopting a somewhat hostile out applications. This is reflected, for Wherever appropriate, this includes a attitude toward fuzzy systems research. instance, in the fact that fuzzy logic brief survey of the principal researchers
brief survey of the principal researchers I was told, however, that the controcontrol was one of the three major and research activities.
versy in Japan is not as severe as in the areas of focus in the original program
West. Some speculations on why this is at the Laboratory for International Fuzzy The Japan Society for Fuzzy the case are taken up in the concluding Engineering Research (LIFE), whereas Theory and Research (SOFT) section of this report. 3 years later, we find that the subject is barely mentioned within its current SOFT was founded and held its first Laboratory for International program. What is perhaps more cor- annual meeting in 1984, at which time Fuzzy Systems Engineering rect, however, is that only an initial it had about 20 members. As of 1991, Research (UFE), Yokohama stage of theoretical development is membership amounts to 1,800 individnow more or less complete, and moving uals and 100 corporations. The current LIFE is a 6-year project (1 April to the next stage will require solving an president is Professor Michio Sugeno
president is Professor Michio Sugeno 1988 through 31 March 1995) funded assortment of substantially more difficult of the Tokyo Institute of Technology by the Japanese Ministry of International Trade and Industry (MITI) in conjunc- an electric power plant; and of some corporation and producing a tion with 49 major Japanese corpora- (c) Process Control Based on Fuzzy series of
series of 8 or 10 publications. The tions. Its stated objectives are (1) to Dynamic Models, Project Leader publications then count in place of a promote research and development Suzuki, system for control of doctoral dissertation. Many of the people (R&D) on applications of fuzzy theory chemical, fermentation, or envi- at LIFE are currently involved in this to engineering and (2) to promote
process. The advantage to the particidomestic and international exchange
pating companies is that their employees on the study of fuzzy theory. The man- 2. Intelligent Robot, all three proj- in this manner gain good research aging director is Toshiro Terano, Pro- ects are oriented toward develop- experience, together with a solid groundfessor of Control Engineering at Hosei ing a “home helper” robot: ing of knowledge in fuzzy technology, University. Professor Terano is well (a) Understanding Language which is then brought back to the corknown as one of the first Japanese Instructions, Project Leader porate research laboratory. researchers in fuzzy systems.
Yokogawa, general problems of The senior researchers at LIFE are At its inception, LIFE was orga- natural language understanding and more experienced corporate employees nized into three "laboratories": (1) Fuzzy communication; (b) Image Under- or university professors who visit LIFE Control (especially for production standing, Project Leader Norita, on a part-time basis and serve mainly processes and robots), (2) Fuzzy Intel- concerned with color and object an advisory role. The institute also lectual Information Processing (deci- recognition and with the interface encourages both short and extended sion support systems, image understand- between images and language; and visits by foreigners. Other important ing, expert system shells, diagnosis (c) Planning and Intelligent Con- functions include organizing national system for power station, language trol, Project Leader Maeda, con- and international conferences and understanding for robots, and evalua- cerned with moving-obstacle avoid- seminars. LIFE has become the de facto tion and understanding of numerical ance and high-level autonomous hub of activity and communications information), and (3) Fuzzy Computer control.
regarding fuzzy systems research in (including system architecture, hard
Japan. ware, and software).
Circumference: In 1990, however, LIFE's goals and (a) Fuzzy Associative Memory, The Science and Technology research emphasis underwent some Project Leader Yamaguchi, fuzzy Agency (STA) ) fundamental changes, largely because neural networks, integrating fuzzy by that time fuzzy technology had found logic with neural net techniques; The STA is roughly the equivalent widespread use in industrial and com- (b) Fuzzy Expert Systems, Project of the U.S. National Science Foundamercial products. Most significantly, it Leader Tano, a general-purposetion (NSF), funding primarily universitywas decided that the subject of fuzzy fuzzy expert system shell, which related research. (STA is also a small control had advanced to sufficient also plays a role in the Decision contributor to LIFE, but only at 1/20 maturity that there was no further need Support System Project (la above); the amount provided by MITI.) In 1989, , for a laboratoryon that topic; it was felt and (c) Fuzzy Computer, Project STA initiated a program titled “Fuzzy that the necessary theoretical work Leader Tokunaga, hardware and Systems and Their Application to was largely complete and that it was software for computers designed Human and Natural Problems," which next up to the corporations to start specifically for fuzzy information began funding fuzzy systems research producing applications. As a result, the processing.
at the rate of around ¥200 million (or projects were reorganized into three
$1.5 million) per year. As such, the groups, having three projects each: In addition to promoting industrial actual level of funding provided by
applications of fuzzy technology, LIFE STA for fuzzy systems research is much 1. Decision Support Group: also plays an important educational smaller than MITI's contribution to
(a) Decision Support System, role. The institute houses about 60 LIFE, but it actually supports a great Project Leader Tano, to develop a researchers, most of whom are younger, deal more research activity. This is fully automated expert system that junior level employees of major corpo- because STA has no need to provide buys and sells currencies on the rations who are pursuing advanced salaries. Japanese university professors international exchange; (b) Plant university degrees. In the Japanese are paid year-round by their instituOperation Support System, Project university system, a student with a tions and are normally provided at least Leader Yoneda, a system for con- masters degree can obtain a Ph.D. by minimal research facilities. Thus grants trolling, monitoring, and diagnosing going to work in the research department from STA cover only funding for
equipment, travel to meetings, business- (15) Application of Fuzzy Logic to inverted pendulum experiment. Conrelated entertainment, books, and other Social and Management Systems Struction on a new building, planned to incidental expenses.
house 40 researchers plus administraDuring the summer of 1991 there (16) Earthquake Forecasting tive personnel, had been scheduled to were 18 projects being funded under
begin very soon and was expected to be this program, and the word at that time (17) Prediction of Air Pollution in completed in 1992. The institute also was that the program's budget was soon
produces the new Journal of the Fuzzy to be enlarged. The list of project titles
Logic Systems Institute, which currently is as follows: (18) Modeling of Plant Growth is published only in Japanese.
In addition to Yamakawa's work on (1) Fuzzy Logic
Fuzzy Logic Systems Institute the fuzzy inference and defuzzifier chips,
(FLSÍ), Kyushu Institute of he has recently developed a fuzzy neuron (2) Algorithm of Fuzzy Reasoning Technology, lizuka, Kyushu chip and demonstrated its utility by
means of an application to pattern (3) Programming Language and FLSI was established in March 1990 recognition. A device has been conArchitecture
to conduct experimental research into structed that correctly recognizes hand
fuzzy information processing and neuro- written characters with the same degree Intelligent Control of High- science and to promote the wider use of accuracy as prior devices, but with Speed and Unstable Systems of the scientific findings in these domains. much greater speed. The results of these (helicopter)
The chairman and person primarily experiments were to be presented at
responsible for its creation is Professor the International Fuzzy Expert Systems (5) Intelligent Control of Ill
Takeshi Yamakawa, whose work on (IFES) conference in November 1991. Structured Systems (fermenta- fuzzy logic controllers was discussed in Other fuzzy systems researchers at tion plant) a foregoing section.
KIT are Toyohiko Hirota and Torao
The initial budget for FLSI was Yanaru, together with their student (6) Real-Time Image Understand- ¥100 million ($750,000) provided by Tomokazu Nakamura. Some of their ing (fuzzy-neuro system) 13 private corporations working in col- work concerns the theory of fuzzy infer
laboration with the Kyushu Institute of ence and its applications in expert sys(7) Recognition of Handwritten Technology (KIT) and Fukuoka Pre- tems. In addition, they have been explorLetters
fecture. From the prefecture's stand- ing the use of projective geometry to
point, this is part of a long-term effort represent various properties of fuzzy (8) Modeling of Sensual Informa- to establish a new center of technolog- systems, a study initiated recently in
tion Processing (image and sen- ical industry in the lizuka area, which the United States by Bart Kosko. sor fusion)
formerly was a coal mining community
and is now economically relatively Omron Corporation, (9) Human Interface in Home depressed. The local campus of KIT Fuzzy Technology Business Automation
was itself established for this purpose, Promotion Center, Kyoto
and Yamakawa's move there from his (10) Human Interface for High-Speed former position at Kumamoto Univer- Omron is primarily a second tier and Unstable Machine
sity was largely to enable his participa- corporation, producing components that tion in the formation of FLSI.
other manufacturers use in products (11) Evaluation of Complex Systems New participants have continued to for the commercial markets. It also
(scenario evaluation, fuzzy case- join, most notably Omron Corpora- makes equipment for large manufacbased reasoning)
tion, which as note earlier has traded turers. Omron invested early in fuzzy
its support in part for the rights to technology and is now the world's lead(12) Fuzzy Information Retrieval Yamakawa's patents. At the time of my ing innovator in the creation of fuzzy
visit, the institute consisted of three logic devices. A brief chronology fol(13) Fuzzy Association (voice researchers, an administrative direc- lows below. Some of this has been recognition)
tor, and two staff assistants, in a small extracted from a 9 May 1991 report by
temporary building near KIT. I was Thomas Hagemann of the German (14) Evaluation of Reliability of given a demonstration there of the National Research Center for Computer
Large-Scale Systems (safety)
Science, Tokyo. Additional informa- A prototype fuzzy chip designed by (1) tracking problems (noisy, timetion has been derived from my own Yamakawa was manufactured (now variant systems), e.g., temperature visit. My hosts were Masaki Arao and the FZ-5000), and the hybrid control, tension control, position Satoru Isaka.
(Fuzzy+PID) temperature controller control, chemical plant control In 1966, Omron developed the ESAF was developed. Omron took DECIVAC, an analog computer that part in establishing LIFE, participated (2) tuning problems (conflicting conimplemented a type of probabilistic in the Fuzzy Committee of the STA, straints), e.g., gain tuning, crane decision making and was to a certain and received a grant of ¥600 million control extent intended to address the same ($4.5 million) from the Japan Research variety of problems as was fuzzy sets. It and Development Corporation (another (3) human factors (feelings, intuitions), is interesting that this work almost STA organization).
e.g., cruise control, engine diagexactly coincided with Zadeh's first In 1989, 10 new fuzzy products were nostic systems, steering control publication in this area.
announced by Omron, and 60 fuzzy In 1983, when the first fuzzy tech- demonstrations appeared at the Omron (4) interpolation (multi-inputs, multinology appeared in Japan (control of a Festival, an idea contest held regularly level processing), e.g., automotive drinking water treatment plant by Fuji within Omron. Professor Zadeh became air conditioner, washing machine, Electric and the Sendai Railway proj- a senior advisor to Omron, and the gas/liquid flow regulator, manuect by Hitachi), Takeshi Yamakawa Fuzzy Project Team was transformed facturing device control, label idenvisited Omron's Tokyo office in search into the Fuzzy Technology Business tification of financial support for his fuzzy inte- Promotion Center, while the team leader grated circuit (IC), a hand-made sam- was dispatched for 2 years as head of (5) classification (complex pattern ple of which he had completed in his one of the original three research labo- recognition), e.g., handwriting laboratory at Kumamoto University in ratories at LIFE.
recognition October of that year. Omron saw the By 1990, Omron's total number of potential for this new technology, and patent applications for fuzzy logic devices The 31 July issue of the Daily Yomiuri in October 1984, Yamakawa came to reached almost 600. These included a newspaper announced an agreement Omron's head office in Kyoto for a fuzzy human body sensor, a fuzzy expert between Omron and NEC“to combine lecture, which was also attended by system for machine diagnosis (together NEC's semiconductor technology with Kazuma Tateishi, the founder of Omron. with the machinery manufacturer Omron's fuzzy logic expertise to develop He showed a deep interest in fuzzy Komatsu), a digital fuzzy chip (the FP- fuzzy logic support systems and microtechnology and that year took control 3000), a fuzzy inference board incor- processors." I was told that Omron of Yamakawa's patent ideas and began porating the digital fuzzy chip (the expects about 30% of its total business developing fuzzy hardware under FB-30AT), and a new tuning method (¥350 billion, or $2.5 billion, in sales Yamakawa's supervision. They also at for fuzzy controllers.
per year) to be fuzzy related by 1995. this time began R&D on fuzzy expert As of my visit in July 1991, the total Other manufacturers are, of course, systems. By 1986, they had produced number of patent applications had also quickly moving into this area, and fuzzy hardware and a medical diagnosis exceeded 700 and included a camera Oki Electric Industry Company has expert system.
that can follow moving objects, a robot recently announced a competing fuzzy The fuzzy boom in Japan began in with sufficient sensitivity to lift cakes inference chip. 1987. As mentioned earlier, this was of tofu, a color combination recognizer, the year Yamakawa presented his a bottle cap recognizer, and a temper- Matsushita Electric Industrial inverted pendulum at the Second ature controller for a chemical reac- (MEI) Company, Central Congress of the International Fuzzy tion plant. As mentioned earlier, plans Research Laboratories, Osaka Systems Association. The chips used in were being laid for developing approxthat demonstration had been built at imately 40 fuzzy logic devices specifi- Matsushita, also known as National/ Omron. Also during this year, Omron cally for use in automobiles.
Panasonic, is one of the world's leading built a “fuzzy computer” (a fuzzy arith- Omron currently employs more that producers of consumer products metic logic unit (ALU)]. In 1988 this 30 engineers working in fuzzy systems employing fuzzy technology. Others was marketed as the FZ-1000, and a R&D, with applications divided into include Fuji Electric, Fuji Film, Hitachi, special task force, the Fuzzy Project 5 problem types:
Mitsubishi Electric, Nissan Auto, Sharp, Team, was established within Omron.
and Toshiba. Through their advertising,