A general observation about controls

Why does it seem like control theory texts go out of their way to obfuscate things?

From naked observer bias, it seems like most stuff you can apply controls to is much more linear than not. System identification seems less complex than it's given credit for.

Obviously, there are cases where this isn't so. But most stuff makes a nice straight line when you apply a perturbation. It's C(1/s)+D plant all over the place.

I like trolling through Laplace transforms and all that, but in the case where you have to communicate with others, it gets in the way. And it seems to slow things down; you end up following all sorts of blind alleys. Some of that is just semi-bad writing, but some of is... something else.

As an unwashed, philistine sort, I have to wonder. Some of it is learning the lingo, some of it is just my propensity for density, I am sure.

Reply to
Les Cargill
Loading thread data ...

I don't do oil refineries or power plants, but in my world, taking any badly behaved bits and wrapping a local proportional loop around them, maybe with a bit of pre-distortion, seems to make everything much faster and better behaved. (For instance, I just did a simple approximate square-rooter to run a small heater, where I couldn't just use PWM as usual due to pickup issues.)

I've noticed that. There are even turf wars, e.g. between Phelan and the whole rest of the community.

There are fields in optics that are like that, especially radiometry/photometry. Beyond stuffy. One suspects there's a sort of masonic brotherhood involved, because they go out of their way to obscure the connections to the rest of the field, and the thicket of stupid nomenclature, overlapping systems of units, and duelling standards bodies makes a simple field complicated.

Cheers

Phil Hobbs

Reply to
Phil Hobbs

Torquer motor with limited current -- if you treat it as linear, then your mechanism will bash into the stop at nearly full speed while somewhere inside the controller there's a differentiator calling for about 100 times as much "stop" as is available from the amplifier.

Motor with friction -- if you treat it as linear it'll either never get close to target, or it'll sit there hunting, or both. (And even with fixes applied it'll never get _to_ the target, and you need to be able to tell the product line manager what the system's capabilities are).

Motor driving a mechanism with backlash -- if you treat it as linear you'll get similar problems as with friction. The details of the wrongness will be different, but the wrongness will remain.

PLL -- linear models are fine once the loop is in lock, but won't help you attain lock quickly or, sometimes, even reliably.

As long as you don't mind the occasional wildly wrong answer that looks right, yes.

True. So you need all the fancy theory for what's left over.

Semi-bad writing, or outright bad reading?

There are a lot of loops that can be closed with good old seat-of-the- pants methods. In fact, I teach those methods (see

formatting link
But the big drawback to these is that when you start you don't know if they'll work for a given combination of plant and goals, if they don't work you won't know why or if something else would work with that plant, and if they do work you don't know how robust the solution is.

There has been more than once when the biggest contribution I could make as a control systems designer was to take a proposed mechanical arrangement, do some modeling, and show why the desired performance goals could never be achieved with that design. You can't do that without a good knowledge of control theory.

Reply to
Tim Wescott

Nor do I.

Yep. Although my experience is that there's usually enough distortion already.

So there's money in it. There should be money in it. However...

Thanks for your thoughts, Phil.

Reply to
Les Cargill

I have yet to have the privilege of dealing with (electric) motors.

That seems to be related to, but not quite the same thing. I've done a few example of "go chase the PLL back out of the woods" and it usually involved a state machine. Basically a statistically controlled, variable state reset.

The FPGA had panic lines when counters overflowed. I just had to do slow things to talk it off the ledge.

That's peripheral to, but not quite the same thing as I'd think of as controls. Sure, there's stuff inside and outside the unit circle. No doubt it's deeply related.

Yeah, the usual. The rules take a week, the exceptions, the rest of your life.

Probably :) No, there are a lot of examples of people who are talking about something but it's difficult to say exactly what.

Feh.

Nice! Thanks for the link.

Yes. And that's why I think the hard part is plant characterization. The problem with *general* texts is that they can't be specific enough to deal with *a* plant.

Thanks much, Tim. I'll keep at it. Might be time for some labs.

Reply to
Les Cargill

The predistortion (e.g. the rough square-rooter) is to linearize the actuator, mainly to reduce the variations in the local loop gain. In this instance, it reduced the gain variation of a heater from about 5:1 down to +-20%, so I can have more loop bandwidth.

Well, job security, anyway. How much radiometrists get paid is a mystery. Some of them dislike me because I'm rude about them in public. To be fair, making good radiometric measurements is hard--it's not their competence I'm impugning, it's their stupid nomenclature and their tendency to try to use standards bodies to make everybody else follow it. (Quality folks (and some safety folks) have the same tendency.)

For instance, they redefined "intensity" to mean something completely different from what everybody else means by it.

Cheers

Phil Hobbs

Reply to
Phil Hobbs

I'm not sure if you're using the term "characterize" the way I might.

If you mean constructing a plant model, for example from mechanical drawings and data sheets (which can be done, often with fair accuracy for a first cut), then I'd call that "modeling". It requires not only a knowledge of control systems (or at least of differential equations), but at least a working knowledge of the system at hand. I can do this for mechanical and electrical systems, but if you tossed me at a chemical problem I wouldn't be very helpful unless you gave me a physical chemist to work with.

I think of "characterizing" as measuring a plant's behavior and coming up with the parameters for a model. In that case you want to at least start with some notion of what the model will be, particularly if there are any nonlinearities to contend with. You can always do a swept-sine measurement (in fact, it's usually my measurement of choice) and design using Bode plots, but even then it's a Really Good Idea to know what nonlinearities may be lurking under the hood.

Reply to
Tim Wescott

PolyTech Forum website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.