| Let's review
how the FFT process
works by examining the following computer generated
signal: |
Figure 1 - Shows Approximately
9 Shaft Rotations (470 msecs)
|
The conventional
FFT process focuses on sinusoids - namely, mathematically calculating what
series of simple sinusoids (signals) were combined to generate the signal
we see here. What can we see from the above plot ?
-
A low frequency
sinusoid that shows about 9 cycles across Fig. 1. That is the 1x rpm signal.
-
Some frequency
modulation of that signal (compare the positive going side of the wave
to the negative going side of the wave).
-
A large number
of spikes, or impacts, that occur across the plot and appear to vary somewhat
in intensity (the size of the spike).
Figure 1 is a
typical example of a plot that an analyst might collect - 9 rotations of
a shaft. But although the 1x sinusoid is fairly clear, the impacts are
not. Let's zoom in a bit. |
Figure 2 - Shows Approximately
2 Shaft Rotations (115 msecs)
|
Cutting the
displayed sample to just over 115 msecs (about 2 shaft rotations), we can
now clearly see:
-
The frequency
modulation of the 1x rpm signal.
-
The ringdown frequency
of the impacts.
-
If we simply count
the number of impacts in one cycle (from 30 - 80 msecs, for instance),
we would find about 4-5 per shaft revolution (or "x RPM").
It should be clear
to us as analysts that this is an impact occurring and investigation of
the period involved (time between impacts) should lead us to a diagnosis.
But more often than not, the analyst will not be using the time domain
- they will be using FFT analysis. What does an FFT performed on this signal
generate ? |
Figure 3 - FFT Generated From
Signal In Figure 1
|
-
1x, 2x and 3x
rpm peaks. These are
probably due to the frequency modulation present.
-
A series of peaks
at high frequencies that are spaced about 5400 cpm apart.
-
The absence
of a peak at or near 5x rpm - the impact frequency. This is because there
is no sinusoidal motion associated with the frequency of the impacts -
only the ringdown frequency that results from the impacts.
But where do the peaks between 31,000
and 65,000 cpm come from ? How does the FFT process come to "see" them
? |
