Intelligent Real-Time Control System (Fuzzy CMAC)
Summary:
With the help of a BMDO SBIR Phase II contract, Intelligent Automation, Inc., has developed a hybrid neural network technology called Fuzzy CMAC that can be used for pattern recognition systems. Fuzzy CMAC combines the flexibility of fuzzy logic-driven controllers with the self-learning capability of neural networks. This technology is employed in a ballistics analysis system, DRUGFIRE®, that is marketed by Mnemonic Systems, Inc. Intelligent Automation hopes to find additional markets in the medical microscopy field for the Fuzzy CMAC technology.
Technology Description:
Intelligent Automation, Inc. (IAI; Rockville, MD), has developed a hybrid artificial intelligence (AI) technology that combines the flexibility of fuzzy logic-driven controllers with the self-learning capability of neural networks. The company predicts this AI technology will simplify the design of real-time control systems and, at the same time, allow control systems to adapt to changing characteristics of the system being controlled. IAI calls its AI technology Fuzzy CMAC because it combines a fuzzy logic approach with a type of neural network called a Cerebellar Model Arithmetic Computer (CMAC).
Current real-time control systems, as used in weapons, manufacturing systems, and scientific instrumentation, have several shortcomings. First, it is very difficult to design a system that accurately models all of the complexities the system may encounter. Secondly, the characteristics of most systems change over time, leaving the control unit ill-suited for the new conditions. Therefore, the next step in the development of control systems may be in the direction of adaptive self-learning. Controllers will be programmed to ''learn'' their changing environments without human intervention. To this end, IAI has developed a technology combining fuzzy logic and neural nets to address this need.
Neural nets are already being employed in real-time control systems. However, according to initial tests, the average learning rates of IAI's CMAC neural networks are at least ten times faster than conventional neural networks. Because the Fuzzy CMAC can learn faster and more accurately than conventional CMACs, it will facilitate higher performance control, plus faster and better adaptation to changing characteristics during operation. For example, a Fuzzy CMAC could help pilots to better control their aircraft by dynamically adjusting the control parameters as the aircraft dynamics change because of fuel consumed, altitude, speed, weight (after weapons have been deployed), etc.
MDA Origins:
In 1994, IAI successfully demonstrated Fuzzy CMAC and developed several software tools for designing Fuzzy CMAC systems. Under a BMDO SBIR Phase II contract, IAI is exploring potential applications in missile control, target classification, and satellite control.
Spinoff Applications:
Because its learning rate and speed are superior to conventional neural networks, Fuzzy CMAC is emerging as a potential new technology for real-time control and pattern recognition. IAI has studied applying Fuzzy CMAC to the following:
•Machine tool chatter control: In a contract with Lockheed Martin, IAI is developing an active machine tool chatter control system for manufacturing applications. The ''brain'' of this system is the Fuzzy CMAC technology, which learns the characteristics of the system. The ''muscles'' of the system are a set of actuators developed by Lockheed Martin. The worth of high precision machine tools is directly proportional to their cutting speed and today, the limiting factor in the cutting speed of such machines is chatter. If IAI's systems can prevent chatter and thereby allow machines to cut 50 percent faster, the machines will be worth 50 percent more than they are today.
•Flight and weapons control: Fuzzy CMAC will make flight control systems more reliable and adaptive, which will enhance maneuverability, accommodate uncertainties, and simplify design. In flight, a pilot pushes the stick, hoping the aircraft will respond exactly the way he or she wants. But a plane may respond differently when it is loaded down with ordinance or has full or near empty gas tanks, etc. The Fuzzy CMAC control system is able to adjust for these differences in real time, allowing the plane to respond to the pilot's control as precisely as possible. In addition, this technology could also be used to help weapons systems lock on to targets faster and more accurately.
•Scanning tunneling microscopy: In a contract with the Army's Space and Strategic Defense Command (Huntsville AL), IAI is using Fuzzy CMAC to develop the next generation of scanning probe microscopes. The Fuzzy CMAC will allow the microscope to learn the characteristics of the sample as it scans, thereby allowing faster and more accurate image capture.
•Active vibration control: Also, in the Army contract, IAI is developing an active vibration control system.
•Traffic management: In a test, IAI has used Fuzzy CMAC to detect abnormal traffic incidents on a highway. The test results were 99% accurate.
•Communications: IAI has shown that Fuzzy CMAC could be used in a non-linear adaptive signal filter.
Commercialization:
Thus far, IAI has had tremendous commercial success employing the Fuzzy CMAC in a scanning device called RotoScan. Rotoscan is the chief component of the DRUGFIRE® system, a ballistics analysis system developed and marketed by Mnemonic Systems, Inc., a wholly owned subsidiary of Nichols Research. RotoScan digitizes the striations (scratches) on bullets caused by the action of the gun barrel on the bullet. These striations are in effect a fingerprint of the gun on the bullet and can be used as evidence that a particular gun fired the bullet. RotoScan has a centering sleeve that positions the bullet in clay with the nose end pointed down. The bullet is then rotated in front of a line scan camera to capture the entire surface of the bullet. A computer then stores the image on a computer disk, which is then used to identify and match bullets. Used by the FBI as well as by domestic and foreign crimelabs, DRUGFIRE has solved over 1,000 crimes and has achieved more than $1.5 million in sales.
IAI is interested in discussing other potential applications with end-users. Using Fuzzy CMAC, the company could develop a broad range of real-time control and pattern recognition systems, including those used in fingerprint identification, flight control, vibration control, and incident detection on highways.
Company Profile:
Intelligent Automation, Inc., is headquartered in Rockville, MD. Founded in 1986, IAI employs over 40 individuals. It specializes in artificial intelligence, imaging, signal processing, integrated circuit design, neural networks, Internet-based training, and robotics.
Contact Information:
Dr. Leonard Haynes
Intelligent Automation, Inc.
7519 Standish Place, Suite 200
Rockville, MD 20855
Tel:(301) 294-5200
Fax:(301) 294-5201
email: lhaynes@i-a-i.com
web: www.i-a-i.com
|