Step Response Analysis

[chi_visor]

hello friends, I am trying to derive a mathematical model of a mechanical system using the open-loop response of the system. I gave a step response at the input and obtained the corresponding output from the mechanical system. I wanted to know whether there is software where i can feed the data and plot the magnitude and phase response. I have also observed that there is a lot of noise in the sampled data. How will it affect the response?

Thanks in advance...

 

[kokabanga]

LabView from national instruments

 

[chi_visor]

hi kokabanga, ive been working with labview to obtain a magnitude and phase but the response has a lot of noise and the final system estimation is not at all accurate. My system is overdamped and ive been looking at a method to fit a curve to the actual response using the equation G(s) = K/((s+a)(s+b)). I'm looking for an iterative algorithm which can do this based on the actual step response data.

 

[dick_freebird]

If you were able to produce an ideal step, then you would know
the true (=ideal) input harmonic content. But of course there is
no such thing. However you could acquire the input edge. Then
FFT / DFT it and you will have a set of X-axis frequency points
and Y-axis divisors.

Acquire the output waveform with an identical channel and FFT
it, and you "should" have the same frequency points, and your
Y-axis numerators. Now you have a magnitude vector.

Take the raw delay (threshold, TBD; small signal, large signal,
are these the same or different and is small signal a valid
assumption-of-convenience?) and this time can be used to
get phase(f) =2pi*t*f (or something like that). Fill the phase
vector with those generated values.

Now many FFT / DFT algorithms can generate a lot of
numerical artifacts if you don't get your window and stuff
right. Needing to stop clean on a fundamental cycle if
there is one, etc. You might have to take an initial pass
to determine appropriate transform settings or something.
But you might also want to simply discard any datapoints
that did not appear in the input FFT (with perhaps a
binning about those centroids to pull in any slightly-offset
but probably-real energy).

Seems complicated but perhaps doable. Not very tractable
for closed form hand analysis. So I wouldn't hold out a lot
of hope for "mathematical". Computational, maybe.