Mocern vs Fuzzy Controls

When I was a university student, I was always drawn to Conventional
Controls, that is working in the frequency domain, and with lag and
lead
compensators, and making pole zero diagrams.  This can also be done
with
digital.


But what was presented as the future was Modern Controls, done in the
time
domain, or the sample time domain for digital.  The rationale was that
you
could get the optimum, you could minimize a given cost function, you
could
minimize the effects of noise, and there was no trial and error, no
heuristic dimension.


While I was attracted to this, I always did have questions about it.
Of
course the heuristic is in how you define the cost function.  Also, I
was
never clear what the applications of this really were.  Making
inherently
unstable aircraft fly, like the FB 117 stealth fighter bomber, or the
closely related project of docking space craft, or running oil
refineries.
Beyond that I couldn't see much.


The examples to justfy the supremecy of Modern Control are things
like
balancing inverted pendulms.  No of course it can do this.  Otherwise
how
could a Saturn V rocket work for the moon shots.  Otherwise it would
fall
apart.   You can't be twiddling knobs with something like that.


In the many years since, the science has advanced, to deal with
nonlinearity
and adapability.  Is the state of the art H infinity?


But something else has happened to.  A new thinking has emerged.  Or
is it
just new buzz words.  I'm speaking about fuzzy logic, or Fuzzy
Controls.


I look at it and I see no theory at all.  Its a total heuristic.   The
main
thing which would make the result different from Conventional
Controls, is
that Fuzzy Control schemes tend to be nonlinear.   But I also know
that such
schemes, with no lag or lead, are the most likely to be unstable.


So can I dismiss Fuzzy as just nonsense hype, like for National Public
Radio
and Ira Flato's Science Friday?    There are so many books about it,
and as
I see it is all nonsense.


I would dismiss it, except that some years back they were making
claims
about doing things Modern Controls could not.  They were claiming to
do 2
and 3 jointed inverted pendulums, and flexible pendulums, the rope
trick.


So what are we to make of this.   I have never seen anyone try to
reconcile
Modern Controls and Fuzzy Controls.


Frank

----------------------------------------------------------------------------------------
Come Join:
http://groups.google.com/group/realtime_signal_and_control

realtime_signal_and_control@googlegroups.com

frank.agee@gmail.com wrote:


> So can I dismiss Fuzzy as just nonsense hype, like for National Public
> Radio
> and Ira Flato's Science Friday?    There are so many books about it,
> and as
> I see it is all nonsense.

An engineer who can find the coefficients of a basic PID controller as 
the result of the theory has a cost of at least $150k p.a.

An engineer who can put to work a basic PID controller by tweaking has a 
cost of $120k.

An engineer who can do the fuzzy-wuzzy genetic neural wavelets in matlab 
so something would finally work somehow has a cost of $100k.

So here we go.


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

Vladimir Vassilevsky wrote:
> frank.agee@gmail.com wrote:
>> So can I dismiss Fuzzy as just nonsense hype, like for National Public
>> Radio and Ira Flato's Science Friday?    There are so many books about 
>> it, and as I see it is all nonsense.
> 
> An engineer who can find the coefficients of a basic PID controller as 
> the result of the theory has a cost of at least $150k p.a.
> An engineer who can put to work a basic PID controller by tweaking has a 
> cost of $120k.
> An engineer who can do the fuzzy-wuzzy genetic neural wavelets in matlab 
> so something would finally work somehow has a cost of $100k.
> So here we go.
> Vladimir Vassilevsky
> DSP and Mixed Signal Design Consultant
> http://www.abvolt.com

Whoa, where did you come up with those numbers.

Fuzzy seems like it has some advantage for low numbers of I/O and nonlinear systems, but when I looked at it 5+ years ago, the tools weren't available to scale to MIMO systems. Maybe that has changed of late.  But for linear or piecewise linear and linear MIMO systems, linear techniques work well.  If you only know how to use a hammer, every problem looks like a nail.

On Sat, 12 Apr 2008 04:15:16 -0700 (PDT), "frank.agee@gmail.com"
<frank.agee@gmail.com> wrote:


>But something else has happened to.  A new thinking has emerged.  Or
>is it
>just new buzz words.  I'm speaking about fuzzy logic, or Fuzzy
>Controls.

Fuzzy isn't new at all.  It's been around at least twenty years, more
like twenty-five IIRC.    The automatic transmission controller in
Saturn automobiles, when the company was brand new in 1990 (or so),
used a Fuzzy Logic control system.   They even used that fact in their
marketing.

The fact that it hasn't caught on too heavily or replaced more
traditional methodologies should tell us something, I think.


>So can I dismiss Fuzzy as just nonsense hype, like for National Public
>Radio
>and Ira Flato's Science Friday?    There are so many books about it,
>and as
>I see it is all nonsense.
>
>I would dismiss it, except that some years back they were making
>claims
>about doing things Modern Controls could not.  They were claiming to
>do 2
>and 3 jointed inverted pendulums, and flexible pendulums, the rope
>trick.
>
>So what are we to make of this.   I have never seen anyone try to
>reconcile
>Modern Controls and Fuzzy Controls.

I think it's just a different design methodology that may have some
advantages when the problem isn't well understood, some
characteristics aren't well understood, or there are some
non-linearities that make more traditional analysis difficult or
impossible.

Some of the people who've gone around evangelizing Fuzzy methods have
done it great harm by mischaracterizing it or comparative systems.  It
seems like the sort of thing that some people latch onto in order to
carve out a niche for themselves as "Fuzzy Experts" or something,
whether or not it really makes sense in a particular application.

More than a decade ago I heard some guy on NPR describing how Fuzzy
was better than a non-Fuzzy digital control system since digital
controls were only either on or off.   He gave an example of an
elevator that would stop abruptly (i.e., the drive would go from "on"
to "off") at the selected floor, jarring the passengers and freight,
while a fuzzy controller would slow down and stop smoothly.

It's that sort of muddled, outright-wrong, downright stupid thinking
that has sometimes gotten associated with Fuzzy, and I don't think
it's necessarily the fault of Fuzzy methodologies.

Nevertheless, I agree that I've not seen a well-considered comparison
of Fuzzy to more classical methodologies, and that sort of thing would
be useful, I think.

Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

"Eric Jacobsen" <eric.jacobsen@ieee.org> wrote in message 
news:fvu104lmctjej6rpq5rnibfpgv1fludf9e@4ax.com...
> On Sat, 12 Apr 2008 04:15:16 -0700 (PDT), "frank.agee@gmail.com"
> <frank.agee@gmail.com> wrote:
>
>
>>But something else has happened to.  A new thinking has emerged.  Or
>>is it
>>just new buzz words.  I'm speaking about fuzzy logic, or Fuzzy
>>Controls.
>
> Fuzzy isn't new at all.  It's been around at least twenty years, more
> like twenty-five IIRC.    The automatic transmission controller in
> Saturn automobiles, when the company was brand new in 1990 (or so),
> used a Fuzzy Logic control system.   They even used that fact in their
> marketing.
>
> The fact that it hasn't caught on too heavily or replaced more
> traditional methodologies should tell us something, I think.
>
>
>>So can I dismiss Fuzzy as just nonsense hype, like for National Public
>>Radio
>>and Ira Flato's Science Friday?    There are so many books about it,
>>and as
>>I see it is all nonsense.
>>
>>I would dismiss it, except that some years back they were making
>>claims
>>about doing things Modern Controls could not.  They were claiming to
>>do 2
>>and 3 jointed inverted pendulums, and flexible pendulums, the rope
>>trick.
>>
>>So what are we to make of this.   I have never seen anyone try to
>>reconcile
>>Modern Controls and Fuzzy Controls.
>
> I think it's just a different design methodology that may have some
> advantages when the problem isn't well understood, some
> characteristics aren't well understood, or there are some
> non-linearities that make more traditional analysis difficult or
> impossible.
>
> Some of the people who've gone around evangelizing Fuzzy methods have
> done it great harm by mischaracterizing it or comparative systems.  It
> seems like the sort of thing that some people latch onto in order to
> carve out a niche for themselves as "Fuzzy Experts" or something,
> whether or not it really makes sense in a particular application.
>
> More than a decade ago I heard some guy on NPR describing how Fuzzy
> was better than a non-Fuzzy digital control system since digital
> controls were only either on or off.   He gave an example of an
> elevator that would stop abruptly (i.e., the drive would go from "on"
> to "off") at the selected floor, jarring the passengers and freight,
> while a fuzzy controller would slow down and stop smoothly.
>
> It's that sort of muddled, outright-wrong, downright stupid thinking
> that has sometimes gotten associated with Fuzzy, and I don't think
> it's necessarily the fault of Fuzzy methodologies.
>
> Nevertheless, I agree that I've not seen a well-considered comparison
> of Fuzzy to more classical methodologies, and that sort of thing would
> be useful, I think.
>
> Eric Jacobsen
> Minister of Algorithms
> Abineau Communications
> http://www.ericjacobsen.org
Have you seen Bob Pease's colunms in Electronic Design?  Bob is an analog 
guru and doesn't particularly like digital computers.  He presents and 
debunks a few examples of Fuzzy Logic hype, but seems to say that it might 
be useful in some highly nonlinear cases.  He also points out that most 
fuzzy logic papers do not describe the sensors used.

http://www.national.com/rap/Story/fuzzylogic2.html
http://www.national.com/rap/Application/0,1570,25,00.html
http://electronicdesign.com/Articles/ArticleID/4915/4915.html

Daniel Abramovitch's paper is also an interesting read.
http://www.labs.agilent.com/personal/Danny_Abramovitch/pubs/fuz_mat.pdf

   Best wishes,
   --Phil Martel 

On Apr 12, 11:15 pm, "frank.a...@gmail.com" <frank.a...@gmail.com>
wrote:
> When I was a university student, I was always drawn to Conventional
> Controls, that is working in the frequency domain, and with lag and
> lead
> compensators, and making pole zero diagrams.  This can also be done
> with
> digital.
>
> But what was presented as the future was Modern Controls, done in the
> time
> domain, or the sample time domain for digital.  The rationale was that
> you
> could get the optimum, you could minimize a given cost function, you
> could
> minimize the effects of noise, and there was no trial and error, no
> heuristic dimension.
>
> While I was attracted to this, I always did have questions about it.
> Of
> course the heuristic is in how you define the cost function.  Also, I
> was
> never clear what the applications of this really were.  Making
> inherently
> unstable aircraft fly, like the FB 117 stealth fighter bomber, or the
> closely related project of docking space craft, or running oil
> refineries.
> Beyond that I couldn't see much.
>
> The examples to justfy the supremecy of Modern Control are things
> like
> balancing inverted pendulms.  No of course it can do this.  Otherwise
> how
> could a Saturn V rocket work for the moon shots.  Otherwise it would
> fall
> apart.   You can't be twiddling knobs with something like that.
>
> In the many years since, the science has advanced, to deal with
> nonlinearity
> and adapability.  Is the state of the art H infinity?
>
> But something else has happened to.  A new thinking has emerged.  Or
> is it
> just new buzz words.  I'm speaking about fuzzy logic, or Fuzzy
> Controls.
>
> I look at it and I see no theory at all.  Its a total heuristic.   The
> main
> thing which would make the result different from Conventional
> Controls, is
> that Fuzzy Control schemes tend to be nonlinear.   But I also know
> that such
> schemes, with no lag or lead, are the most likely to be unstable.
>
> So can I dismiss Fuzzy as just nonsense hype, like for National Public
> Radio
> and Ira Flato's Science Friday?    There are so many books about it,
> and as
> I see it is all nonsense.
>
> I would dismiss it, except that some years back they were making
> claims
> about doing things Modern Controls could not.  They were claiming to
> do 2
> and 3 jointed inverted pendulums, and flexible pendulums, the rope
> trick.
>
> So what are we to make of this.   I have never seen anyone try to
> reconcile
> Modern Controls and Fuzzy Controls.
>
> Frank
>
> ----------------------------------------------------------------------------------------
> Come Join:http://groups.google.com/group/realtime_signal_and_control
>
> realtime_signal_and_control@googlegroups.com

I think H infinity control has re-invented old-fashed lag-lead control
in a sense. That is, an experienced control engineer will come up with
just as good if not better control system through experience as an H
infinity controller. This is only for SISO systems - not for
multivariable though. The multivariable control of say a jet aircraft
needs more than just a few Bode plots.

L2 minimisation ie LQG etc never lived up to expectation. You needed
to add integral action for starters but it could be made to work. I
doubt it was as good as a good lag-lead controller. (or PID).

Fuzzy logic is for academics with time on their hands and nothing
better to do. I would go for adaptive H infinity if I had a choice but
I see H infinity more as a design tool than an adaptive algorithm.

Inverted pendulums are easily stabilized using PID type algorithms.
You need lots of phase advance and at least one integrator - sometimes
two.

Fuzzy logic has its place I am sure, but controller design would not
be top of my list.


K.

On Sat, 12 Apr 2008 19:55:41 -0400, "Philip Martel"
<pomartel@comcast.net> wrote:

>> Nevertheless, I agree that I've not seen a well-considered comparison
>> of Fuzzy to more classical methodologies, and that sort of thing would
>> be useful, I think.
>>
>> Eric Jacobsen
>> Minister of Algorithms
>> Abineau Communications
>> http://www.ericjacobsen.org
>Have you seen Bob Pease's colunms in Electronic Design? 

Doesn't everybody read his columns?   He's the Amazing Randi/Snopes of
the electronics world.  ;)

>Bob is an analog 
>guru and doesn't particularly like digital computers.  He presents and 
>debunks a few examples of Fuzzy Logic hype, but seems to say that it might 
>be useful in some highly nonlinear cases.  He also points out that most 
>fuzzy logic papers do not describe the sensors used.
>
>http://www.national.com/rap/Story/fuzzylogic2.html
>http://www.national.com/rap/Application/0,1570,25,00.html
>http://electronicdesign.com/Articles/ArticleID/4915/4915.html
>
>Daniel Abramovitch's paper is also an interesting read.
>http://www.labs.agilent.com/personal/Danny_Abramovitch/pubs/fuz_mat.pdf
>
>   Best wishes,
>   --Phil Martel 
>

Thanks for the links, those seem to be to the point.

Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

kronecker@yahoo.co.uk wrote:

   ...

> Inverted pendulums are easily stabilized using PID type algorithms.
> You need lots of phase advance and at least one integrator - sometimes
> two.
> 
> Fuzzy logic has its place I am sure, but controller design would not
> be top of my list.

The inverted pendulum is long-since solved problem. I first saw one in 
operation in the late 1940s, at the IRE show in the Kingsbridge Armory 
in The Bronx. Small DC servos weren't practical then. The standard servo 
was a two-phase motor with high-resistance windings. The excitation of 
one winding was fixed, and the other, driven in quadrature, applied the 
control by varying amplitude and polarity. That was a better match to 
vacuum tubes' high voltage and low-current. (The fixed excitation was 
effectively a carrier, and the control was AM suppressed carrier. We 
knew how to to build carrier-based lead-lag networks.)

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

On Sat, 12 Apr 2008 04:15:16 -0700, frank.agee@gmail.com wrote:

> When I was a university student, I was always drawn to Conventional
> Controls, that is working in the frequency domain, and with lag and lead
> compensators, and making pole zero diagrams.  This can also be done with
> digital.
> 
> 
> But what was presented as the future was Modern Controls, done in the
> time
> domain, or the sample time domain for digital.  The rationale was that
> you
> could get the optimum, you could minimize a given cost function, you
> could
> minimize the effects of noise, and there was no trial and error, no
> heuristic dimension.
> 
> 
> While I was attracted to this, I always did have questions about it. Of
> course the heuristic is in how you define the cost function.  Also, I
> was
> never clear what the applications of this really were.  Making
> inherently
> unstable aircraft fly, like the FB 117 stealth fighter bomber, or the
> closely related project of docking space craft, or running oil
> refineries.
> Beyond that I couldn't see much.
> 
> 
> The examples to justfy the supremecy of Modern Control are things like
> balancing inverted pendulms.  No of course it can do this.  Otherwise
> how
> could a Saturn V rocket work for the moon shots.  Otherwise it would
> fall
> apart.   You can't be twiddling knobs with something like that.
> 
> 
> In the many years since, the science has advanced, to deal with
> nonlinearity
> and adapability.  Is the state of the art H infinity?
> 
> 
> But something else has happened to.  A new thinking has emerged.  Or is
> it
> just new buzz words.  I'm speaking about fuzzy logic, or Fuzzy Controls.
> 
> 
> I look at it and I see no theory at all.  Its a total heuristic.   The
> main
> thing which would make the result different from Conventional Controls,
> is
> that Fuzzy Control schemes tend to be nonlinear.   But I also know that
> such
> schemes, with no lag or lead, are the most likely to be unstable.
> 
> 
> So can I dismiss Fuzzy as just nonsense hype, like for National Public
> Radio
> and Ira Flato's Science Friday?    There are so many books about it, and
> as
> I see it is all nonsense.
> 
> 
> I would dismiss it, except that some years back they were making claims
> about doing things Modern Controls could not.  They were claiming to do
> 2
> and 3 jointed inverted pendulums, and flexible pendulums, the rope
> trick.
> 
> 
> So what are we to make of this.   I have never seen anyone try to
> reconcile
> Modern Controls and Fuzzy Controls.
> 
AFAIK, fuzzy control was a reaction to the complexity and ambiguities of 
'modern' control.  Zadeh was looking for a systemic way to express 
heuristic rules, and he succeeded.  Where the folks who study fuzzy logic 
go astray, IMHO, is in wildly over-applying fuzzy logic and in assuming 
that one can solve problems _just_ by using fuzzy logic and heuristics 
without ever going back and rigorously analyzing the controllers one has 
generated using fuzzy logic to encode ones heuristics.

Given that Zadeh was _trying_ to supplant modern controls, I don't know 
that you can reconcile the two.  If you can reconcile two such differing 
world views, you could do the world a lot more good if you'd head over to 
Jerusalem and talk sense into the Palestinians and Israelis.

'Modern' control (if it's called 'modern' it was invented between 50 and 
100 years ago -- think about that) provides some very good analysis tools 
of MIMO systems, but it doesn't provide very good tools for synthesizing 
controllers.  Sure, you can use pole placement techniques if you don't 
mind your system being unstable in the face of plant variations, or you 
can use optimal control techniques if you don't mind your system being 
unstable in the face of plant variations.  If you _do_ mind your system 
being unstable in the face of plant variations you can try to use robust 
control design techniques and hope that you've correctly captured all of 
your plant variations, and that the resulting plant description won't 
blow little bits of your system model all over mathemagic land when you 
jam it into the analysis engine.

In my work I end up using three basic approaches to control system 
design.  Most often I'll design by frequency-domain analysis (good ol' 
Bode and Nyquist plots).  I'll often start with a model built from first 
principals, but if at all possible I'll end up using a model derived from 
measurements.  Next most often, if I'm working with a really easy plant 
and a problem that doesn't call for pushing the system performance very 
close to the limits of what's possible I'll do some seat-of-the-pants PID 
loop tuning (this will also sometimes segue into measurements, and more 
formal analysis and design).  Finally, about twice a decade or so I'll do 
a design using robust control design techniques, if I have a plant that I 
know is going to be varying all over the map and if I know that stability 
is a real concern.

About the only place that I end up using something that looks like 
'modern' control design is if I'm implementing a controller on a DSP I'll 
take advantage of the fact that it's way faster on a DSP to multiply a 
vector by a sparse matrix in a loop than it is to only multiply the 
significant terms together; consequently it's often most time efficient 
to implement ones controller as a state-space controller with 
accompanying matrix multiply than it is to do it as a chain of PID, 
notch, and lead-lag elements.

-- 
Tim Wescott
Control systems and communications consulting
http://www.wescottdesign.com

Need to learn how to apply control theory in your embedded system?
"Applied Control Theory for Embedded Systems" by Tim Wescott
Elsevier/Newnes, http://www.wescottdesign.com/actfes/actfes.html

"frank.agee@gmail.com" wrote:

> I would dismiss it, except that some years back they were making
> claims about doing things Modern Controls could not.  They were claiming to
> do 2 and 3 jointed inverted pendulums, and flexible pendulums, the rope
> trick.
>
> So what are we to make of this.   I have never seen anyone try to
> reconcile Modern Controls and Fuzzy Controls.

I can make some comments as someone who has used fuzzy logic in
control systems and have developed tools for fuzzy logic
implementations.

 I would still probably choose a classic control system for a static
control system.

We spent about two years looking looking at fuzzy logic based
control implementations one that could compete with conventional
implementations. Once we found one then the list now seems to
be endless.

The early examples we found were AC motor start-up controllers for
a rolling mill and aircraft flight control systems (Fuel burn in flight changes
the system characteristics and dynamics)

Fuzzy control systems often involve a lot more parameters than
conventional control systems. A home environment controller my have
rules that account for time of day and transition between rule sets
combining the results.

My experience is that fuzzy implementations win in non linear
systems, systems whose control requirements change over time
or systems where several different competing controls are in place.
Many fuzzy based systems are less computationally intensive
than conventional alternatives.

Fuzzy logic has lost a lot of marketing  hype but is heavily used
in consumer applications. Some of the current applications that
I am currently aware of in consumer products are, video
stabilization in camera's and television sets, cooking profiles
in micro wave ovens, toasters, rice cookers and industrial ovens,
and home environment control.



Regards,

--
Walter Banks
Byte Craft Limited
1 (519) 888-6911
http://www.bytecraft.com
walter@bytecraft.com