Is this valuable to build such a platform?

I am really sorry to bother you on this topic, but I am nearlly
brought down to such a tricky problem, please share me some experience
and guidance..

No.  Tell your advisor to do it first. If he can't do then he should
be replaced.
Why must the manipulator be flexible?  This makes the project MUCH
more difficult.
Ihe design should be as stiff as possible.
It can be done but not easily/cheaply.  Controlling a flexible
maniplator is beyond a college project.
I am not saying it cant't be done.  It will take too much time and
effort when the correct solution is to keep the manipluator design
stiff.
I would fire the advisor.  Show this to your advisor.

I dug up your old post
http://groups.google.com/group/sci.engr.control/browse_frm/thread/e70506a4ee=
da5c1b/09ff3efcdc99e460#09ff3efcdc99e460
You are not ready for this.  I doubt you instructor is.
I can also see I missed a post where I got flamed.

Peter Nachtwey

On Sat, 19 Jan 2008 19:21:03 -0800, workaholic wrote:


In what way?

No model can be exact; trying to make an exact model is a guarantee of
failure.

No matter what you do, your model must capture the salient points of the
system in such a manner that your results mean something.  Most
importantly, the salient points that you capture must be the _right_ ones
for the question _at hand_ -- you may find yourself modeling the same
system in many different ways as you try to choke various different
answers out of the math.

For example, if you can create a model of the manipulator that
demonstrates that it'll never work right then you shouldn't have to go
any farther.  Basically, if you have a (hopefully simple) model that
models a better mechanism than you can ever build, and show through
analysis that such a mechanism just won't work, you've shown that the
upper performance bound is lower than the minimum acceptable.

Conversely, if you create a model of the manipulator that demonstrates
that it can always work, you can proceed with confidence.  Here you would
want a (hopefully simple) model that models a mechanism that you can
always better in practice, and show through analysis that the lower
performance bound is acceptable.

So ask yourself (and your adviser) just what exactly you're trying to
achieve with your model, and aim to make it do that, and not too much
more.

--
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

This is true

This is pure BS and you know it.

Peter Nachtwey

On Sun, 20 Jan 2008 02:12:54 -0800, pnachtwey wrote:


A: No model can be exact (you agreed to this).
B: Failure happens when you try to do something and don't succeed
   (this is the commonly accepted definition, I think?)

If A is true, then no attempt to exactly model a real-world process can
possibly succeed, because no model is exact.

Ergo, any try at making an exact model guarantees failure.

It's not BS, it's just a logical restatement of the original statement
(with which you agree).

Fortunately we're engineers, not mathematicians, so we know (or should
know) that we don't _need_ an exact model, we just need a model that's
good enough.

That was my point in my original response -- if you try for an "exact"
model you will fail.  I have seen (and done in my younger years) attempts
to make a model "exact" -- one either works on it for an indefinite
period of time and then stops, or one gets a model that fits everywhere
except in one's blind spots.

--
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

Thanks for everybody, you help me to make things clearer, since my
advisor, as a mathematican,  is not an expert in mechanics, who is
just assigned to a task of superivising a lab doing this...

If a relatively exact model is not quite reasonable, what if I
establish one on a commercial simulation platform? There are some
brand new software can help to do this, however, it seems to be
unacceptable by the academics.

So it seems that I have to do some theoretical research instead of
this...

BTW, what the material of the Canada's Arm? Where can I get some
specific information about it$B!)(BAre there any related papers?

Best regards,

Jie

On Sun, 20 Jan 2008 23:28:06 -0800, workaholic wrote:


There's a world of difference between "exact" and "relatively exact".  
Getting a model that's close enough ("exact enough") for a wide range of
purposes is probably doable -- but only if you start with a notion of
what those purposes are, and what effects you can leave out and what you
must leave in.  This is _not_ something that will necessarily be
successful if you just go at it in a hit or miss fashion.

I'm not sure what commercial simulation platform you're talking about.  
At best, a simulation tool is going to insulate you from the knowledge
and experience you're supposed to be developing; at worst you'll be
depending on someone else getting it right, and my experience is that
where such tools don't have outright bugs they can be very poorly
documented.

Whatever mathematical model you develop, it should be just that: a
mathematical model.  Were I your adviser I'd be looking for something of
the form

x(t) = f(x(t), u(t), t), y(t) = g(x(t), u(t), t)

where x(t) is your system's state vector (hopefully it's a finite
collection of discrete states), u(t) is the system's input (both command
and disturbance), t is time, y(t) is the system output, f is a memoryless
state evolution function, and g is a memoryless output function.  You
_should_ be able to describe just about any real dynamic system with the
above pair of equations.  Analyzing it is another problem, but you should
be able to _describe_ it that way.

--
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

On Sun, 20 Jan 2008 23:28:06 -0800, workaholic wrote:


Another thought:

Is your adviser the only only control theorist on campus?  This sort of
modeling is the kind of thing that you would expect a guy with a PhD in
mechanical engineering, with a specialty in control, would be able to
help you out with.

In the US it would be appropriate to find out who he or she is, and show
up at his/her office to politely ask for help -- then thank him/her later
with a six pack of Really Good Beer.  In the US a university would be
considered pretty dysfunctional if no such cross-discipline help were to
occur.  If you're too shy, you may consider forwarding the notion to your
adviser and asking if they'd do the asking for you.

--
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

Thanks for your guidance, I am fortunate enough to have a Ph.D student
in my lab, who got a bachelor's degree in ME. But since his opinion is
always reverse to my advisor's, I am not sure who's suggestion I
should choose...

It looks to me that the Phd with the ME degree, Bruce and I think this
is unrealistically difficult.   Tim's idea of asking for help may not
work because the professors may not be able to help.   I know I don't
ask professor's for help.
Modeling in one dimention is hard enough.  Look through this use group
and see how much information there is on modeling.  Not much.  You
need to pick something to do that is outside your comfort zone but
just barely.    How about moving a pendulum from a over head carriage
or something like that.  The idea would be to get the pendulum to stop
at the destination without swinging.   This has practical application
in crane operation.

Peter Nachtwey

It seems that little progress have been on flexibla manipulator, since
I have seen bunches of papers on IEEE journals and transactions. Much
of them are aiming at such arm in one dimension.

Sometimes researchers are prone to utilize different control scheme,
which is often advanced and obscure, on a different plant, instead of
solving realistic questions...

Aside from what's already been mentioned, that this project is probably
unrealistically difficult, I get the impression that you may need help with
setting up the project itself. That includes clearly defining the
objectives, activities (including how long things will take)and resources,
and structuring the whole thing as a series of linked tasks.

Your supervisor has a responsibility to guide you through this, it's
something that you learn by doing. This wouldn't be the first time that I've
come across supervision on academic projects that is possibly both ignorant
and negligent in regard to mentoring inexperienced people in the project
management aspects. If Mr C is still alive and reading this, you're a good
example!!