The Oxymoron of Control Systems: Theoretical, Yet Broadly Practical
|
4.23.03 -- No one who knows Bob Bitmead would describe him as a business-as-usual type of bloke. But, even so, my discussion with him to learn about his research surprised me by the way it started: He was asking the questions, thank you. For example: What did sugar cane crushing mills, speech compression systems for telephones, jet engines, and cell phone geolocation all have in common?
He had me there.
It turns out that control systems for all four applications derive from the same mathematical principles. I knew we were in for a ride now…
Control systems, you say.
"The research we control systems folks do," says Bitmead, a professor of Mechanical and Aerospace Engineering in the UCSD Jacobs School of Engineering, "is fundamental and generalizable, rather than specific like much scientific work. The techniques I develop and work on from a theoretical perspective allow me to think of many applications problems in which to test them."
Bitmead began with the example of controlling an unfamiliar shower. "Because of the danger in getting scalded," he says, "you tend to change the temperature slowly from cold to hotter." If you understand the delay, he says, you can set the shower head up properly to get much higher performance, quicker response, if you will. But you need to understand the dynamics of the system.
"The rubber really meets the road in control systems," explains Bitmead, "when you consider how to manage the shower temperature in the face of measured or unmeasured disturbances like toilet flushes, other taps being turned on or off, etc. This side of control systems is called disturbance rejection and, after stabilization, accounts for much of the derived applications benefit."
According to Bitmead, lots of problems have similar characteristics, especially simple physical things like filling a gas tank or understanding transport delay in a telecommunications network. Control systems theory is of particular value as it appears to be the key in achieving high performance in these kinds of complex systems.
Bitmead gives another example: "If you've ever talked by satellite phone, to my homeland of Australia, for instance," he says, "the time delay makes it next to impossible to communicate with the other person unless you agree to a different protocol. This is a good example of how dynamics affect the usefulness of a given system. We've learned to cope by using simple spoken cues, like saying, literally, 'you talk first' and indicating when you're done talking by using some accepted code words." If the dynamics of the system were better understood, system performance could be improved, which would eliminate the need for such behavioral changes.
A self-described "boundary dweller" among aerospace, mathematics, and electrical engineering, Bitmead explains that he studies and develops mathematical formulations of system models, then compares model output with experimental data to refine the models in an iterative process.
"The fundamental thing we work on is stability of feedback control systems. Let's return to the shower as a good example. If you overcorrect for hot water, you can create a situation where temperature swings become wilder and wilder - the opposite of stability in which the temperature swings become less and less until you home in on a desirable temperature. What we study is the math underlying the dynamical model of the system as extracted from measured data."
According to Bitmead, his work may sound esoteric but, in fact, it's incredibly practical. "The contribution of control systems work," he says, "can be fundamental mathematical theorems that have fairly immediate and dramatic applications in a wide variety of fields."
How about some more examples?
Bitmead turned to work he has done with the Australian steel industry, particularly as related to hot strip mills, which turn out steel for use in refrigerators, car panels, and soft drink cans. One such project was with BHP Co. Ltd, a very large company working in steel, petroleum, and minerals.
BHP's mills move slabs of steel (40 tons, 12 m long x 2 m wide x 250 mm thick) into a furnace where they're heated from ambient temperature to about 1250 degrees Celsius. Then the steel is "rolled" seven times back and forth through a reversing roughing mill, which reduces the thickness of the steel from 250 mm to 25 mm. At this point the length has increased from 12 m to 120 m. Then the steel takes a single pass through the finishing mill, which reduces it to 3 mm in height but 1.2 km in length!
"The problem we dealt with was how to control the furnace," explains Bitmead. "If we got that right, the final product would have much higher value because the steel would have more uniform characteristics." The goal was for the temperature throughout the slab to be 1250 degrees.
Bitmead measured the temperature inside the slab by measuring the force needed to deform the steel in the reversing roughing mill. But he also had to take into account air cooling, water cooling, work heating (as the rolling adds energy to the slab), and work hardening (as steel is deformed, energy is added, which changes the crystal structure so it becomes harder). From the measurements of force in the rolling mill, he was able to measure the temperature of the core of the steel. From that point, it was relatively simple to control the furnace to produce uniform temperature in the steel.
In this project, underscoring the tangible value of his work, Bitmead delivered a product that provided steel temperature estimates to company operators. "So, in the end," he says, "the control system was a piece of software." Though clearly a computational scientist, Bitmead sees computers as communications, rather than computing, devices. "It's not about computing power but doing something mathematically alert with computers," he says. You need the right math to write the right software to control the systems.
Or take another example: An emergency call is placed on a cell phone, and emergency responders need to determine the location of the caller. The cell phone is in the field. It has sensors and an antenna to communicate with various base stations that are in the vicinity and managed by a handoff management system that can also tell you where the phone is located. "The math we use to reconstruct the location of the phone," says Bitmead, "is strongly related to the math we use to reconstruct the core temperature of the steel."
Bitmead and I agree that control systems are still hard for the layperson to understand. "The message I have the most trouble conveying," says Bitmead, "is that a model that will be used for control design is different from one used for prediction or simulation. When I'm developing a model from which to design a controller, it's often the case that a simple model is more powerful than a reductionist model that attempts to capture infinitesimal details by understanding all the subsystems."
The latter seems to be the philosophy of the aerospace industry, which believes that, to get higher performance, it's necessary to understand more and more detail. Bitmead's approach, on the contrary, is to look harder at the math model. It's impossible to model all the detail, he maintains, so deriving enormously complicated models and assuming they approximate the truth is "wrong-headed." According to him, control systems design can be based on just a few key properties. "This seems to be well understood in Europe where entire departments are devoted to this study," says Bitmead shaking his head, "but not in the U.S."
Bitmead is the recent recipient of an endowed chair by Cymer in high-performance dynamical systems modeling and control. This is an obvious honor, says Bitmead, but what he finds particularly exciting is that it shows a high-tech company that produces lasers for photolithography sees modeling and control as enabling its corporate future.
"The really fundamentally rewarding feature of the work I do is that the applications drive the math," he says. "The best math I've worked on in control systems," he explains, "has come about because my attempts in practical problems have failed using existing techniques."
The work Bitmead does need not, or should not in large measure, be immediately applicable, he concludes. "We want to work on technology for the next decade," he says emphatically, "rather than just for tomorrow. The Jacobs School's long-term vision is evidenced by the fact that it's chosen to invest in this area."