Prev: toeplitz matrix
Next: Finding turning points in a curve
From: Wayne King on 22 Feb 2010 08:17
"Madhu Shurpali" <madhu.shurpall(a)cvgrp.com> wrote in message <hltp77$nrj$1(a)fred.mathworks.com>...
> > Hi Madhu, In addition to Rajiv's fine advice (System Identification Toolbox has a lot of nice tools), can you please give us your sampling frequency? Without your sampling frequency, nobody can get the right normalized frequencies to try and help you.
> > Wayne
>
> Wayne and Rajiv thanks a lot for the advice.
>
> Wayne: I think the Sampling freq doesnt come into picutre here. The way I got my Complex Freq response is as follows:
>
> I input a displacement of 0.5 Hz freq and 5 mm amplitude sine wave into my specimen and measured the force as the output. Now my specimen has some damping in it so when I plot Disp Vs Force I get a elliptical hysteresis curve. From the Hysteresis curve I get the stiffness and damping. I did this at Freqs of 0.5 Hz, 1Hz, 2HZ, 4Hz and 6Hz keeping the amplitude constant. Thus I got the stiffness and damping at each Freq. From the stiffness and damping at each freq I came up with the Complex Freq response.
>
> Also since I am trying to get the transfer function in the s domain, freq normalisation will not be done.
>
> Thanks for all the help

Sorry Madhu, I read invfreqz instead of infreqs, my mistake. Do you obtain the transfer function based on

freq = [0.314159 3.141592654, 6.283185307, 12.56637061, 25.13274123, 37.69911184];
Real = [31.93 32.914 34.25 34.6 33.18 36.57 ];
Imag = [3.446 2.941 1.828 1.646 7.213 7.955];
h = Real+1j*Imag;
[b,a]= invfreqs(h,freq,3,4);
sys = tf(b,a);
% yields
Transfer function:
-1.065e004 s^3 - 1.564e005 s^2 - 6.269e006 s - 7.586e006
--------------------------------------------------------
s^4 - 264.7 s^3 - 4039 s^2 - 1.768e005 s - 2.558e005


Can you say what doesn't seem right about it? You say looking at it, it doesn't seem right.

Sorry again about misreading your post.
Wayne
From: Madhu Shurpali on 22 Feb 2010 08:27
> Sorry Madhu, I read invfreqz instead of infreqs, my mistake. Do you obtain the transfer function based on
>
> freq = [0.314159 3.141592654, 6.283185307, 12.56637061, 25.13274123, 37.69911184];
> Real = [31.93 32.914 34.25 34.6 33.18 36.57 ];
> Imag = [3.446 2.941 1.828 1.646 7.213 7.955];
> h = Real+1j*Imag;
> [b,a]= invfreqs(h,freq,3,4);
> sys = tf(b,a);
> % yields
> Transfer function:
> -1.065e004 s^3 - 1.564e005 s^2 - 6.269e006 s - 7.586e006
> --------------------------------------------------------
> s^4 - 264.7 s^3 - 4039 s^2 - 1.768e005 s - 2.558e005
>
>
> Can you say what doesn't seem right about it? You say looking at it, it doesn't seem right.
>
> Sorry again about misreading your post.
> Wayne

Ok, heres what I did: To check my model in time domain, I built a simple simulink model consisting of the following:
A sinewave input block, a transfer function block and a scope block. Output of sinewave goes into tansfer function block and output of transfer function goes into scope. I define the transfer function as per the coefficients I got from the invfreqs function. For solving I used a variable step, ode45 solver with other parameters at the default value.

Now when I simulate, with a 10mm, 5Hz sinewave I get unrealistically high values. I expect a sinewave with a gain of about 34. I am able to understand this phenomena.

Thanks for the help.

Madhu
From: Madhu Shurpali on 22 Feb 2010 09:03
Sorry for the typo in the previous post. It shoul dbe like the following:

"I am unable to understand the phenomena"
From: Madhu Shurpali on 23 Feb 2010 06:40
"Madhu Shurpali" <madhu.shurpall(a)cvgrp.com> wrote in message <hlu2qm$3ve$1(a)fred.mathworks.com>...
> Sorry for the typo in the previous post. It shoul dbe like the following:
>
> "I am unable to understand the phenomena"

Somebody please help.. I am not able to go forward from here.
From: Oskar Vivero on 23 Feb 2010 08:53
"Madhu Shurpali" <madhu.shurpall(a)cvgrp.com> wrote in message <hm0eqn$kp1$1(a)fred.mathworks.com>...
> "Madhu Shurpali" <madhu.shurpall(a)cvgrp.com> wrote in message <hlu2qm$3ve$1(a)fred.mathworks.com>...
> > Sorry for the typo in the previous post. It shoul dbe like the following:
> >
> > "I am unable to understand the phenomena"
>
> Somebody please help.. I am not able to go forward from here.

While using the system identification toolbox is good advice, there are some things you need to check before that, for example:

1) Noise levels, if big your estimates will be biased and hence your response won't look alike to your physical plant.
2) Why are you picking such model orders in the first place. If you over or under specify your transfer function structure your estimates will also be wrong. In the same vein, and if the noise levels are small there is no reason why invfreq shouldn't work unless your model structure is incorrect.

I hope this helps
First  |  Prev  |  Next  |  Last
Pages: 1 2 3
Prev: toeplitz matrix
Next: Finding turning points in a curve