[From Bruce Gregory (990406.1147)]
"_A thermostatically controlled heater or furnace automatically
regulating the temperature of a room or enclosure_ is a control system.
The input to this system is a reference temperature, usually specified
by appropriately setting a thermostat. The output is the actual
temperature of the room or enclosure."
_Feedback and Control Systems_ Distefano, Stubberud, and Williams
I have no trouble with calling a reference level an input, but calling
the controlled variable an "output" certainly leaves something to be
desired in terms of clarity--output of what? Omitting any discussion of
how the thermostat "knows" what the "output" is leaves much to be
desired. How do they get anything to work?
I'll know never to discuss control theory with an engineer.
Bruce Gregory
[From Bill Powers (990406.1005 MDT)]
Bruce Gregory (990406.1147)--
I have no trouble with calling a reference level an input, but calling
the controlled variable an "output" certainly leaves something to be
desired in terms of clarity--output of what? Omitting any discussion of
how the thermostat "knows" what the "output" is leaves much to be
desired. How do they get anything to work?
I'll know never to discuss control theory with an engineer.
For about 40 years I've harbored the idea that "real control engineers"
know so much more about control than I do that I have to be very cirumspect
in making claims or drawing conclusions. This has been largely because I
could never make myself sit down and really learn LaPlace transforms and
all the rest of that --er -- stuff. My whole career in PCT has been devoted
to developing simple methods of analysis that give correct answers without
getting into what for me is difficult mathematics.
But I've also been uncomfortable about what I find in servomechanism
textbooks, used, presumably, to acquaint engineers with control system
design and analysis. If the student is lucky, there is an introduction and
maybe a first chapter that give a general overview of control systems,
showing a couple of simple examples like a flush toilet or a temperature
control. Then in Chapter 2 the student is plunged right into transform
methods or frequency-domain computations in the complex plane. If this
represents what the students are actually taught, I don't see any way that
new control engineers can develop any feel at all for how control systems
really work. I'm reminded of that nice Israeli control engineer who was on
CSGnet for a while, six or seven years ago. His explanation of how a
thermostat works was nothing short of bizarre. He simply couldn't accept
that control systems control only what their sensors tell them, not what
the remote environment is actually doing. He finally signed off the net,
saying he just couldn't change his way of thinking.
After the wierd encounter with "modern control theory" I began to think
that the suspected lack of fundamental understanding of control among
control engineers was real. But the only reason this peculiar new
conception of control could have gained acceptance was that the old
approach was vulnerable. Engineers were not taught about reasl control
systems in the old school any more than in the new school. For most
engineers, control system design was a rote procedure, whether they were
taught in terms of shifting poles and zeros around or in terms of Kalman
filters.
All of that is probably just a wrong impression gleaned from some
unfortunate encounters with a few control engineers. But maybe not. Maybe
the approach I've worked out is really something engineers ought to learn
before they start getting into mathematics that is too abstract to help the
intuition. It's not that I object to the mathematics -- but a good
mathematical analysis can't make up for a poor initial conception of a system.
Thanks for pulling the trigger.
Best,
Bill P.
[From Bruce Gregory (990406.1310 EST)]
Bill Powers (990406.1005 MDT)
All of that is probably just a wrong impression gleaned from some
unfortunate encounters with a few control engineers. But
maybe not. Maybe
the approach I've worked out is really something engineers
ought to learn
before they start getting into mathematics that is too
abstract to help the
intuition. It's not that I object to the mathematics -- but a good
mathematical analysis can't make up for a poor initial
conception of a system.
This problem shows up in learning physics as well. "Good" students
master the mathematical manipulation, but most have little ability to
work with the concepts. Thus, even at the best schools "plugging and
chugging" is the approach of choice. Many faculty decry this situation
but have no idea how to change it. I think this failure stems from an
inadequate (I'm being generous here!) model of learning. But that's
another story...
Bruce Gregory