Calculating Operational Deflection Shapes

[DNgineer87]

I am trying to use Labview to obtain Operational Deflection Shapes, but the magnitude of the imaginary plots is wrong. 

I have triax accelerometer data from multiple locations on a structure.  I'm taking the Autopower of the responses and the crosspower of the responses and references.  I then use the magnitude of the Autopower, and the phase of the cross power to create a complex array.  I then convert to the complex data to Real and Imaginary for My ODS plots.  With this technique I get amplitudes that are much too low or much to high. 

Has anyone done this in Labview and would you be willing to share the details?

[Ian_Ren]

The unit of autopower is EU^2. Did you take the square root of the autopower spectrum?

Ian 

[dsb@NI]

What is the excitation to your structure? Does it make sense for you to divide the magnitudes by the magnitude of the reference autospectrum? Could the unexpected amplitudes be an issue of units? As Ian asks - a matter of taking the square root. Are the computed ODS amplitudes in terms of accelerations or displacements, and are these the same units as your expected values? One last question, do you have multiple references? How are you combining the ODS from these references?

Message Edited by Doug Bendele on 10-31-2005 04:56 PM

[DNgineer87]

Thanks for your replies.

I'm using units of g.  I had taken the sqrt of the imaginary data, but your message made me realize I must take the sqrt of the magnitude instead.  That brought the imaginary plots to within a factor of about 1.5 of what MEscope calculates.

I'm applying a Hanning window and its window correction factor is 1.5.  With all these adjustments I am within about 10% of MEscope's values.  I'd like to be closer so I can use them more or less interchangably for data analysis and comparison.

I'm recording 6 triax accels simultaneously (two references and four response locations) and using a "for" loop to index the arrays to generate FRFs for each response/reference combination.  My understanding from Vibrant Technology and LDS literature is that ODS FRFs are composed of the magnitude of the response autopower, coupled with the phase of the reference/response crosspower. 

As I understand it, the only time the reference autopower magnitude is used is if the data is not all collected at the same time.  Then the reference autopower from one data set to the next is used to "normalize" the FRFs.

I plan to investigate what windowing correction MEscope uses.  I also wonder if the FFT routines or other mathematics within the two programs are causing the disparity in results.

Did you compare your ODS FRF results from Labview with any other vibration software?  Did they compare well? 

Any other suggestions?

[Ian_Ren]

10% sounds better, which is less than 0.5dB difference.

Did you use the same data file for both MEScope and your LabView program, and did you use same overlap and average for autopower and cross-power calculation?

Ian

[dsb@NI]

A factor of 1.5 and then still off by 10% - could it instead be a factor of 1.41 (sqrt of 2)? This factor might lead to more direct comparisons between the implementations and might very well be necessary as ODS results would typically be returned in peak units (g pk) while the LV FFT returns rms units (g rms). You do not need to apply a window scale factor to the spectra if you apply the window via the FFT Power Spectrum VI in the LV Waveform Measurements palette. If you want to post your LV code, we can probably provide more directed suggestions.

[DNgineer87]

I suspect the difference may be the averaging routine and/or the digits of precision used by each program.

I have attached the heart of the VI.  If you run it in without averaging the result at 30 Hz is slightly greater than the MEscope results.  The vector or RMS averaging cause the results to be significantly lower.  The linear averaging causes the imaginary plot to be "one-sided" (no positive values).  Only the exponential averaging gives a +/- symmetric plot (as it should look).

I've studied the "ma_spectrum averaging (complex).vi" inside the crosspower VI and it appears to expect the 1-D array to already be divided into blocks for averaging.  I can't see how it subdivides or overlaps segments of the data.

[DNgineer87]

see reply to Ian

[dsb@NI]

I could not run the VI in the attached LLB as it is missing subVI ODS FRF re-Im.vi. I was surprised to see time-domain filters as these filters will each exhibit a transient response that should be excluded from the analysis. Instead, you can integrate in the frequency domain in order to achieve the desired units.

Without more information about the settings (especially averaging) you have selected in ME'Scope, it is hard to give good guidance about the settings you should use in your LabVIEW code.

I will send you an email so that we can work on this more directly.

[dsb@NI]

My email is doug.bendele@ni.com. Email me if you would like me to help investigate further. I think we may be able to make faster progress if we can share info more quickly (either by email or phone).