FFT in diadem

[xzhcong]

Hello,

I found some information from the DIAdem help and DIAdem manual as follows.

"Note  DIAdem calculates the FFT to powers with a base of two and therefore might not use all the measurement data. Example: If a time signal has 340 values, DIAdem only uses the first 256 (2⁸) values for the FFT."

"DIAdem calculates a FFT for the entire length of the specified channels, even if the channel length is not a power of two."

I don't understand which one is right?

[Otmar]

Hello xzhcong,

They are both right, but I agree this is a little misleading.

The original FFT calculation did require a power of two number of values, but it has since been replaced with the improved DFT calculation (and can be switched back to the original algorithm with a configuration setting in DIAdem).

The setting can be changed from the "Settings" menu in DIAdem, in the Settings > Options > General dialog under "Compatibility"

That setting Specifies whether DIAdem uses the improved FFT algorithm. If the variable has the value TRUE, DIAdem uses the new algorithm.

Description:

Note If the variable has the value FALSE, DIAdem uses the FFT algorithm from DIAdem 10.0 and earlier. DIAdem calculates the FFT of the older FFT algorithms to powers with a base of two and therefore might not use all the measurement data. Example: If a time signal has 340 values, DIAdem only uses the first 256 (28) values for the FFT. Only use the older FFT algorithms if you want to compare your results with earlier data. The functions ChnFFT1, ChnFFT2, and ChnInverseFFT including the related dialog boxes FFT with one time signal, FFT with two time signals, and Inverse FFT are affected by these changes.

The default setting for this variable is True, so the new algorithm will be used unless you specify otherwise.

Please check the attached PDF for more details about the FFT calculation ...

Let me know if you have additional question,

      Otmar

Specifies whether DIAdem uses the improved FFT algorithm. If the variable has the value TRUE, DIAdem uses the new algorithm.

Definition	UseNIFFT, Boolean variable

Note If the variable has the value FALSE, DIAdem uses the FFT algorithm from DIAdem 10.0 and earlier. DIAdem calculates the FFT of the older FFT algorithms to powers with a base of two and therefore might not use all the measurement data. Example: If a time signal has 340 values, DIAdem only uses the first 256 (28) values for the FFT. Only use the older FFT algorithms if you want to compare your results with earlier data. The functions ChnFFT1, ChnFFT2, and ChnInverseFFT including the related dialog boxes FFT with one time signal, FFT with two time signals, and Inverse FFT are affected by these changes.

[xzhcong]

Thank you for your information. I read the attachments you give. I am puzzled about: " A measurement with 1000Hz and 1000 values results in 500 frequencier of 0-1000 Hz -- twice the frequency resolution as before." 0-1000Hz. Is is right ? I think it should be 0-500Hz.

[Ulrich_Bierwisch]

Hello xzhcong,

you are right. It should (and will) be 0-500Hz.

Thanks for your help

Ulrich Bierwisch

[xzhcong]

Hello Otmar D. Foehner,

   I would like to calculate the RMS of vibration acceleration.

   As the difference of "One-sided interval width as % of the channel length" would result in the different RMS result, as in the attachment. So I would like to know what value will fit for my test data? How to decide it?

second, I would like to know about RMS calculation at the begin and last value.

Could you describe them by formula?

Thank you!

Mr.Xu

[elgato23]

Hello All,

I have some additional questions about FFT in NI DIAdem that I was hoping you all could shed some light on.

I am doing some FFT calculations on random vibe data. The data is in the form of G’s peak. Here are some questions:

1) When should random attenuation and period attenuation be used?

2) I want the PSD curve (G^2/Hz). My input data is in G peak. Is the g in the output (g^2/HZ) in the form of RMS, Peak, or Peak to Peak?

3) What is the difference between Arithmetic and exponential averaging? What is recommended for analyzing random vibration data?

Additional info: I am using a Hanning Window and an overlap of 50%. I vary the number of intervals depending on how long/large the TDMS file is.

[Ulrich_Bierwisch]

1) Random ist mathematically correct most of the time. Peak is better is you want to read the peak of sine waves in the signal. For typical PSD-calculations peak ist correct.

2) PSD is based on the RMS calculation.

3) I think, in your case arithmetic is better. This depends on what you are looking for and what the units are. If you have a value of 1 and a value of 2 halve of the time, the average can be (1+2)/2 or sqrt((1^2 + 2^2)/2) if you are looking for the energie in the signal.