model identification

I am trying to identify the friction model of the system. There are several model of friction taking the streabeik effect into
consideration. suppose i choose any model. for example say Tustin model, how do i find its parameters. Well while capturing the experimental data(for a given force measure velocity), the streabeik effect is there only near zero(very small velocity) region. How do i define this low velocity. At lower velocity(where velocity is obtained by numerical differentiation of position), the signal noise ratio is low. Can you kindly help me out with some suggestion or pointer to some useful literature where i can get more insight into this. kristo
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Krish wrote:

Bwa ha ha ha ha ha ha ha!
Oh, er. Sorry.
"The Control Handbook", William Levine, Ed., 1996, CRC Press, has a good article on friction modeling and compensation. I also have a note on my website -- http://www.wescottdesign.com/articles/Friction/friction.html -- which is more about tried-and-true methods of controlling motors rather than any attempts to accurately model friction.
If you have a "normal" motor control problem to solve I _strongly_ suggest that you just design your controller to operate in a friction environment, and that you make it as robust as possible. Friction is a very slippery effect to pin down (pun intended); you'll find that even if you can characterize it exactly today it's sensitive to so many factors that it'll be different tomorrow. It's best to just design your system to be robust to it's vagaries.
The nice thing about all that is it frees you from having to accurately model it -- just design your controller with a lot of margin, and be done with it. And keep in mind that while I'm giving this advise I have spent days sweeping a system with a control systems analyzer to get it's _exact_ linearized transfer function, because that _was_ something that was fairly well controlled, and could be profitably used in design.
--

Tim Wescott
Wescott Design Services
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Karl-Johan Astrom wrote a more academic paper on friction http://www.control.lth.se/~kja / I think the web-page by Tim is nice.
Niclas

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Niclas Granqvist wrote:
You can do something simple or perform complex designs for friction. My experiences, regardless if you do something dirt_simple, or complex, is to monitor the temperature, so and to ensure that your system is not headed for catastrophic failure. Temperature monitoring is very simple, and can be use for testing the implementation as well as for PM (preventative maintenance) sheduling. Things with friction, fail, it's just a question of when they fail, not if.
James

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Niclas Granqvist wrote:

From skimming the paper it appears to be similar to the one in the Control Handbook -- I'll have to read it in depth when I get the chance.
--

Tim Wescott
Wescott Design Services
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

The Astrom work is later than the Control Handbook and describes the now popular LuGre (for Lund-Grenoble) model for friction.
Fred.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Thanks to all. my question was more on identification..so that i can have a simulation study of the practical system. i read some literature on idetification of friction but i am not convinced, they sound me too academic work. kristo
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Well, friction is a nonlinear phenomenon and that means that you will have to work some more. Is this a school project or an application? 1) Gather the data. Also gather validation data to try the model on. 2) Basically you need to settle on a few model structures to try out. 3) Identification is not hard with todays fast PC:s. A lot of people buy matlab and use sophisticated tools. If you only have a few parameters then you can do with much less. I have survived with a C-compiler and a couple of for-loops+numerical integration and trying all possible combinations and then saving the interesting results. The PC could be running a day, but it doesn't matter. Eventually the correct results come out. 4) When you have the parameters then plot model input/output in time and look with your own eyes if it seems to work. If you are not happy go to 2)
Niclas

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here. All logos and trade names are the property of their respective owners.