[Vibration data] Setting FFT line resolution

I am running an FFT to analyze raw vibration data recorded on an engine.

My FFT settings are as follows :

• Output = Amplitude
• Calculation type = Overall (Averaged)
• Window = Hanning
• Resolution Df [Hz] = 25
• Amplitude type = peak
• DC cutoff = None
• Overlap = 90%
• Weighting = Lin
• Averaging type = Peak

However, if I change Line Resolution, the output FFT changes. e.g. lower line resolution results in lower FFT amplitude across the frequency range.

Please can someone explain why this is? I thought it would just affect accuracy, but it fundamentally changes results. Should I also be changing another FFT parameter?

For reference, I use the DEWE 43A DAQ module to record data.

Thank you in advance.

Dear Alexandre,

changing the line resolution can affect the accuracy of the measured data. What was the changed value you used for the line resolution? Could you provide a data file with the mentioned signal so we can look into it?

Regards

I tried several different values of Df (25, 10, 5, 1 Hz). Increasing resolution / reducing Df always leads to lower maximum amplitude calculated by the FFT for each frequency.

File size is much too large to upload, unfortunately.

Can you please explain to me the principle behind setting line resolution value, and how to know I can trust the FFT... if I am required to have no vibration >20g for 50-4000Hz, for example, I can increase the line resolution, resulting in lower FFT which now meets the criteria. Clearly, there has to be some logic behind how to properly do this.

Dear Alexandre,

attached are the instructions on how to upload a big data file. You can only add the channel which you use with the FFT analyser.

Regards

Company restrictions prevent me from uploading any data, I'm afraid. Thank you for your understanding.

Could you please still try to answer my questions above?

Many thanks,

Alex

Dear Alex,

this is actually a normal behaviour for the FFT calculation, which you can also expect in other programs. Here is one example and explanation of the behaviour in Matlab.

We also made a little test with the 10*sine(50)+10*sine(60) wave. When we use 50Hz resolution the main amplitudes are combined in one peak - so the amplitude is higher because there are more "values" of the representative frequency. With the 1Hz resolution we have two apparently separated peaks, but wits smaller amplitude. The peaks are smaller because there are less "values" of representative frequency .

Thank you for your example and the link, they is very helpful!

My next question is - which line resolution should I choose? I am unsure what factors I must consider when doing this. Thank you.

Dear Alex,

I would say it depends on the measurement. If you need exact frequencies from your measurement you should use higher resolution. The result also depends on the Window you choose.

Please look into our Pro training for FFT spectral analysis.

Regards