controller design: Frequency domain or state space approach?

If I want to learn controller design, which is better to learn, do it
in
state space (pole placement, etc..) or use frequency domain (transfer
function,  PID, etc.. classical approach?).

What is done in practice, i.e. real world?

I know for MIMO, one has to use state space. But for SISO, which
approach would you choose?

thank you,
Steve

On Nov 13, 9:22&#4294967295;am, steve nospam <steve.nosp...@gmail.com> wrote:
> If I want to learn controller design, which is better to learn, do it
> in
> state space (pole placement, etc..) or use frequency domain (transfer
> function, &#4294967295;PID, etc.. classical approach?).
>
> What is done in practice, i.e. real world?
>
> I know for MIMO, one has to use state space. But for SISO, which
> approach would you choose?
>
> thank you,
> Steve

Learn stae space. Far more versatile, and contains
'classical' approach as a sub-subject.

Rune

On Sun, 13 Nov 2011 07:07:21 -0800, Rune Allnor wrote:

> On Nov 13, 9:22&nbsp;am, steve nospam <steve.nosp...@gmail.com> wrote:
>> If I want to learn controller design, which is better to learn, do it
>> in
>> state space (pole placement, etc..) or use frequency domain (transfer
>> function, &nbsp;PID, etc.. classical approach?).
>>
>> What is done in practice, i.e. real world?
>>
>> I know for MIMO, one has to use state space. But for SISO, which
>> approach would you choose?
>>
>> thank you,
>> Steve
> 
> Learn state space. Far more versatile, and contains 'classical' approach
> as a sub-subject.

Inneresting.  'cause my take on it is to learn the classical approach, 
because it works well in almost all cases, you can often do your "system 
ID" with sine sweeps (dead easy, and fairly easy to accommodate mild 
system nonlinearities), and it makes a nice mental framework from which 
to springboard to state-space.

As far as I can tell, there are three ways (well, four) to do "modern" 
control design:

1: Use naive pole placement, and fail because your system is 
insufficiently robust.

2: Use pole placement, and succeed because you've been doing it for years 
on similar plants and you know how to fake a robust design.  
Characterized by the statement "but I only choose target poles that'll 
work".

3: Use robust design techniques (I.e. H-infinity) that take parameter 
variation into account, and which are based on mathematical methods that 
take a considerable amount of work to wrap your brain around.

4: Learn how to type in Matlab or Scilab, and use the toolbox functions 
for H-infinity controller design that you should only attempt when you 
understand the math.  This is the "million monkeys with typewriters and 
we're waiting for Othello to pop out" solution, except that there's only 
one monkey and you're it.

-- 
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

On Nov 13, 9:22&#4294967295;pm, steve nospam <steve.nosp...@gmail.com> wrote:
> If I want to learn controller design, which is better to learn, do it
> in
> state space (pole placement, etc..) or use frequency domain (transfer
> function, &#4294967295;PID, etc.. classical approach?).
>
> What is done in practice, i.e. real world?
>
> I know for MIMO, one has to use state space. But for SISO, which
> approach would you choose?
>
> thank you,
> Steve

Learn classical control first - always. Then state-space. State space
has no handle on things like bandwidth which is very important.
If you go H infinity then you can kind of mix the two. No point in
trying to design a controller if you don't understand phase-margin
though.


Hardy

On 13 Nov, 18:28, Tim <t...@seemywebsite.please> wrote:
> On Sun, 13 Nov 2011 07:07:21 -0800, Rune Allnor wrote:
> > On Nov 13, 9:22&nbsp;am, steve nospam <steve.nosp...@gmail.com> wrote:
> >> If I want to learn controller design, which is better to learn, do it
> >> in
> >> state space (pole placement, etc..) or use frequency domain (transfer
> >> function, &nbsp;PID, etc.. classical approach?).
>
> >> What is done in practice, i.e. real world?
>
> >> I know for MIMO, one has to use state space. But for SISO, which
> >> approach would you choose?
>
> >> thank you,
> >> Steve
>
> > Learn state space. Far more versatile, and contains 'classical' approach
> > as a sub-subject.
>
> Inneresting. &#65533;'cause my take on it is to learn the classical approach,
> because it works well in almost all cases,

The intro chapter in Friedland's book on control theory,

http://www.amazon.com/Control-System-Design-Introduction-State-Space/dp/0486442780/ref=sr_1_1?ie=UTF8&qid=1321206179&sr=8-1

contains some rather interesting comparisions between the 'classical'
western approach to contro theory, and the Russian approach that
started
from the state space POV.

After having read that chapter, I will never tout the 'classic'
approach
ever again. It seems it is merely the lazy jack-of-all-trades-but-
master-
of-none's approach to the subject.

Yes, it works well a lot of the time, but when the going gets though
the state space approach - according to Friedland - comes out on top.

Rune

On Sun, 13 Nov 2011 09:48:16 -0800, Rune Allnor wrote:

> On 13 Nov, 18:28, Tim <t...@seemywebsite.please> wrote:
>> On Sun, 13 Nov 2011 07:07:21 -0800, Rune Allnor wrote:
>> > On Nov 13, 9:22&nbsp;am, steve nospam <steve.nosp...@gmail.com> wrote:
>> >> If I want to learn controller design, which is better to learn, do
>> >> it in
>> >> state space (pole placement, etc..) or use frequency domain
>> >> (transfer function, &nbsp;PID, etc.. classical approach?).
>>
>> >> What is done in practice, i.e. real world?
>>
>> >> I know for MIMO, one has to use state space. But for SISO, which
>> >> approach would you choose?
>>
>> >> thank you,
>> >> Steve
>>
>> > Learn state space. Far more versatile, and contains 'classical'
>> > approach as a sub-subject.
>>
>> Inneresting. &#65533;'cause my take on it is to learn the classical approach,
>> because it works well in almost all cases,
> 
> The intro chapter in Friedland's book on control theory,
> 
> http://www.amazon.com/Control-System-Design-Introduction-State-Space/
dp/0486442780/ref=sr_1_1?ie=UTF8&qid=1321206179&sr=8-1
> 
> contains some rather interesting comparisions between the 'classical'
> western approach to contro theory, and the Russian approach that started
> from the state space POV.
> 
> After having read that chapter, I will never tout the 'classic' approach
> ever again. It seems it is merely the lazy jack-of-all-trades-but-
> master-
> of-none's approach to the subject.
> 
> Yes, it works well a lot of the time, but when the going gets though the
> state space approach - according to Friedland - comes out on top.

Indeed it does -- but why use a lathe when a pocket knife will do 
splendidly?

-- 
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

On 13 Nov, 21:38, Tim <t...@seemywebsite.please> wrote:
> On Sun, 13 Nov 2011 09:48:16 -0800, Rune Allnor wrote:
>
> > Yes, it works well a lot of the time, but when the going gets though the
> > state space approach - according to Friedland - comes out on top.
>
> Indeed it does -- but why use a lathe when a pocket knife will do
> splendidly?

That was not the question. The OP asked where to invest
his efforts.

Just out of curiousity: If you, present-day TimW, were given
the opportunity to rewind the calender 20 or 30 or whatever,
years, and give your younger Self and advice as he were to
choose his first class or course on control theory, what would
you tell him? "Take the plain vanilla class now and struggle with
state space later on, amongst running your company and
handling a family"? Or "get to grips with the hard stuff
right away, now that you are young and committed to
studying"?

Rune

steve nospam wrote:

> If I want to learn controller design, which is better to learn, do it
> in
> state space (pole placement, etc..) or use frequency domain (transfer
> function,  PID, etc.. classical approach?).

It doesn't matter. Learn something. Eventually it will make sense.

> What is done in practice, i.e. real world?

In practice, controllers are tweaked. There are too many unknowns and 
intractables.

> I know for MIMO, one has to use state space. But for SISO, which
> approach would you choose?

A highly scientific method of Ziegler and Nichols ?


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

On Sun, 13 Nov 2011 13:08:55 -0800, Rune Allnor wrote:

> On 13 Nov, 21:38, Tim <t...@seemywebsite.please> wrote:
>> On Sun, 13 Nov 2011 09:48:16 -0800, Rune Allnor wrote:
>>
>> > Yes, it works well a lot of the time, but when the going gets though
>> > the state space approach - according to Friedland - comes out on top.
>>
>> Indeed it does -- but why use a lathe when a pocket knife will do
>> splendidly?
> 
> That was not the question. The OP asked where to invest his efforts.
> 
> Just out of curiousity: If you, present-day TimW, were given the
> opportunity to rewind the calender 20 or 30 or whatever, years, and give
> your younger Self and advice as he were to choose his first class or
> course on control theory, what would you tell him? "Take the plain
> vanilla class now and struggle with state space later on, amongst
> running your company and handling a family"? Or "get to grips with the
> hard stuff right away, now that you are young and committed to
> studying"?
> 
> Rune

I would say "take the basics first so that you can understand the fancy 
stuff when you come to it".

My understanding of state-space control was very much informed by my 
understanding of classical control theory, just as my day-to-day use of 
transfer functions for characterizing SISO systems is very much informed 
by my understanding of the state-space reality that underpins any system 
that happens to be single-input, single output.

Classical design methods work quite well when you have a SISO (or even 
"few-input, few-output") systems that are difficult to characterize from 
first principals (e.g., the belt-drive system that I'm working with right 
now for a customer), but which are amenable to swept-sine or other 
relatively easy frequency-domain measurement techniques.

State-space design methods work quite well when you have a system which 
is easy to characterize from first principals well enough to meet your 
performance goals.  This includes systems that tend to vary, 'meta 
systems' where you want to write a reusable controller to work with a 
class of systems, systems that have distinct nonlinearities that render 
any frequency-domain testing difficult to interpret, etc.

Since I use both state-space and classical methods quite profitably, it's 
going to be hard to talk me out of one or the other -- just as you'd have 
a hard time of talking me out of either my lathe or my pocket knife.

-- 
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

On Sun, 13 Nov 2011 15:31:55 -0600, Vladimir Vassilevsky wrote:

> steve nospam wrote:
> 
>> If I want to learn controller design, which is better to learn, do it
>> in
>> state space (pole placement, etc..) or use frequency domain (transfer
>> function,  PID, etc.. classical approach?).
> 
> It doesn't matter. Learn something. Eventually it will make sense.
> 
>> What is done in practice, i.e. real world?
> 
> In practice, controllers are tweaked. There are too many unknowns and
> intractables.

Or they are really, really dumbed down as far as the performance that 
they deliver vs. the performance that their particular plant _could_ 
deliver.  And even then they're tuned and tweaked, sometimes.

>> I know for MIMO, one has to use state space. But for SISO, which
>> approach would you choose?
> 
> A highly scientific method of Ziegler and Nichols ?

Even Ziegler and Nichols were up front about their rules-based tuning 
being ad-hoc, and IIRC they state that their tuning is "good for most 
industrial plants" or some such -- in other words, they pulled those 
numbers out of their donkey*.

* I still don't understand that figure of speech.

-- 
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com