05/07/2008
APPARENT LOW FREQUENCIES IN THE
VIBRATION SPECTRUM
Richard Burton
A fairly frequent question heard from vibration analysts is: “What is
causing the vibration at this low frequency peak in my Spectrum?” The frequency
in question will be very low, typically between 100 CPM and 300 CPM, and will
not appear to relate to any calculated frequency of the physical or electrical parts
of the machine. (See Figure 1)
Figure 1 – Mysterious Low Frequency
This mysterious frequency is actually not real. It is a product of Spectrum
‘ski slope’ and the low corner filter that is being applied to the Spectrum.
There are several common reasons for low frequency amplitude.
However, most of the reasons are due to error, such as a rocking mount,
insufficient settling time, cable connection problems, etc. Very seldom will low
frequency amplitude be due to real motion. The most common reason is math
error, especially when using a general purpose (GP) accelerometer. A GP accel
measures amplitudes to within ± 5% accuracy for frequencies down to roughly
300 CPM. The accuracy degrades rapidly until it is ± 50% at roughly 120 CPM. In
addition, the calculated amplitudes of the first few lines in an integrated FFT
Spectrum have a potential for enormous math error or noise. In fact, the first line
is almost entirely math error, and useless for analysis. Any sensor or math error
present can be grossly magnified in an integrated or double integrated Spectrum,
until it is larger than the actual vibration present in the remainder of the
Spectrum! (See Figure 2) This will not only artificially inflate the overall value, it
will cause fluctuations in the vibration trends, preventing accurate analysis of the
actual machine condition.
Figure 2 – Math Error ‘Ski Slope’
A low corner (high pass) filter is commonly used in order to block this
artificial amplitude in the beginning of the Spectrum. However, this filter is not a
perfectly square filter, as seen in Figure 3. It slopes at an angle, blocking more
and more amplitude as the frequency gets lower (gray area). The standard ‘rule
of thumb’ to minimize ski slope is that the low corner filter for integrated
Spectrums should always be larger than the Spectrum Resolution (width of each
line). While this rule works in most cases, it will not always prevent these ‘ghost’
frequencies from appearing in the Spectrum.
Figure 3 – Low Corner (High Pass) Filter
When this filter is applied to a Spectrum that has a ski slope region
extending beyond the low corner filter setting, it cannot completely remove the
ski slope region. It will therefore cut off the inflated low frequency region in the
middle of the slope, creating an ‘apparent’ or ‘ghost’ frequency in the Spectrum at
very low frequencies. This phenomenon can be seen in Figure 4.
Figure 4 – Ghost Frequency Created by the Filter and Ski Slope Data
It is very important to note that this frequency does NOT actually exist in
the vibration data. It is produced when the low corner filter cuts off the low
frequency ski slope.
If this low corner ‘ghost’ frequency appears in the Spectrum, the analyst
must determine the source of the ski slope in the data. While equipment and/or
math error is the most common source of these errata, random impacting and/or
actual low frequency motion (such as structural resonance or motion) can also be
a source for this problem. The analyst should collect and compare a Spectrum
and Time Waveform (TWF) using the exact same settings. A settling or
temperature strain problem will cause the TWF to appear ‘banana’ shaped. A
rocking base, random impacting or cable problem will also be readily visible in
the TWF.
Another test to see if the problem is math error or real motion is to change
to a low frequency accelerometer with a sensitivity of at least 500 mV/g. (Note: A
low frequency accelerometer will usually require a much longer settling time
before accurate data can be taken.) If the problem is not real motion, the ski
slope will either be significantly reduced or eliminated. If the vibration is still
present, the low frequency accelerometer should have enough amplitude
accuracy to allow the analyst to properly diagnose the source of the vibration.