SUPERCOMPUTERS IN COMMAND: A LOOK AT THE FUTURE
In the February issue we discussed the 1993 Santa Fe Chaos conference. Reader response to that columns was quite high, and we'll continue to keep an eye on chaos theory and how it relates to industrial control. Careful study of the characteristics of chaos and control can help solve those sticky applications whose variables inhabit that nebulous region that lies somewhere between linear relationships and truly random events.
In the mid- '80s, I was doing some work on very-high-speed, rule-based systems for an outfit located inside the Washington beltway. That's not unusual - radical technologies are the life blood of the defense establishment. Some associates, however, learned of the government work and suggested that these "chaotic" technologies be used to solve problems in discrete manufacturing. Here, at Flavors Technology, we were incredulous. Manufacturers - we thought - were only interested in incremental improvement. But times had changed.
It's possible to see why the change occurred. In the days of wooden ships and iron men, all control was done with relays. As electromechanical control systems become more complex, the rows of cabinets containing relays reached length of 50 feet or more. The invention of the programmable controller (PLC) was the means to shrink those cabinets back down to five feet.
But solving that cabinet problem allowed developers to further increase control system complexity. The rows of cabinets began to lengthen again, this time filled not with relays, but with PLCs and computers.
We're convinced that supercomputing power and the science of emergent computation will be the means to shrink those cabinets once again. We sold the business of supplying the defense guys and are concentrating our efforts in the industrial arena.
The computing model in a plant usually has five distinct layers, from the information processing centers to the work floor. We want to reduce those five layers to three: the processing center, the plant host and the super controller.
We knew that use of massively parallel, modern computer architecture's would allow adequate power to be brought to bear. And in a single stroke, we could shrink the node count from 1,000 to 10. But you'd think that a plant configured like that would have to have software of almost bone-crushing complexity. We had to convince our champions at General Motors that the ultimate argument - the Japanese had brought a unit. GM was convinced.
Production hardware and software were delivered in 1991. A high-level rule-based language was used (English) and a decent graphical user interface was installed. Using chaos/complexity technology and supercomputer architecture's reduced hardware and software by greater than an order of magnitude.
A printout of the programming that had been used for the application was a stack of paper 10-1/2 inches high. The first cut by GM in programming, based on our system, reduced the stack by a factor of 13. Since then, we've continued to improve, so that the program is now only several pages long.
How is this done? The architecture is software-centered. We use a multiple-instruction, single-data model, with direct connection to the system's minicomputers and PLCs. One memory is accessed by many thousands of rule-based agents. Each agents has only about three rules. The agents support most mathematically describable functions. PID, fuzzy logic, and neural networks all work well. Inputs and outputs are directly imaged into the memory, with from 100 to 15,000 agents installed over time. With a simple set of rules boiling up from the bottom, the control system evolves from within the computer archtecture.
We have, in effect, a real-time game of factory CIM that is fun to play. We, and others, are also tackling problems in power nets, composite materials, semiconductor design, economics, and dynamic scheduling -- all in real time.
The software is seductive. The power is intoxicating and the technology is real. It is exciting to be a technologist in these times. Come join us.
As appeared in Manufacturing Systems Magazine April 1993 Page 38
References - Table of Contents
Send mail to email@example.com for more information.
|R. Morley Incorporated
614 Nashua Street, Suite 56
Milford, NH 03055-4992 USA
Tel: 603-878-4365 FAX: 603-878-4385