Frequency domain measurements and analysis have
become increasingly popular to diagnose a particular machine fault. This measurement mode
relies on processing the transducer output signal using Fast Fourier Transform (FFT)
algorithms to display the signal amplitudes as a function of frequency. FFT processing
essentially separates complex signals into individual components having a single frequency
content. This type of display is commonly termed a spectrum.
An enhancement of spectral
analysis is to define specific frequency ranges to perform band analysis. Conceptually,
band analysis is similar to filtering a signal. The "filter" searches for
frequencies only within its frequency range. Certain permanently installed machine
monitoring systems offer this capability. This feature is quite effective, once the
particular spectral range and resolution has been determined, to rapidly diagnose machine
faults.
Spectrum
A spectrum display is a display of signal amplitudes on the vertical axis and the signal
frequencies on the horizontal axis. The frequency axis units may be in hertz (hz) or in
cycles per minute (CPM). Hertz or cycles per second may be converted into CPM by
multiplying by 60. For example: 10 hz = 600 CPM. The horizontal axis is scaled from 0 to
some maximum frequency (Fmax). Individual signal frequencies will appear as peaks or
spikes, each having a specific amplitude. Properly setting the Fmax will ensure that all
of the input signal is being analyzed. This setting can be verified mathematically by
summing the square of each amplitude peak. The square root of this summation should
approximate the overall amplitude level obtained directly from the transducer's output.
Band Analysis
 Band
analysis involves selecting frequency ranges of interest to allow rapid determination of a
machine's condition. Generally, each machine fault will generate a specific, unique
frequency as the condition deteriorates.
A band is essentially a band pass filter
allowing only the frequencies with the selected range to be measured. All other
frequencies are excluded from analysis. Many modern machine monitor systems are capable of
monitoring specific frequency ranges using band analysis.
Resolution
 Frequency resolution is an area requiring considerable attention. If the resolution is
inadequate the entire analysis process could be meaningless or incorrect. Some instrument
specifications list the spectral resolution as lines. A high resolution would be 3200
lines per spectrum and a low resolution would be 100 lines per spectrum.
Each line of resolution can be viewed as a bucket or pail of a specific size.
The signal frequencies can be viewed as a tennis ball. If a tennis ball's frequency
matches the frequency range of the bucket, it is placed in the bucket. As the bucket fills
with tennis balls the peaks on the spectrum display rise. Should the frequency range of
the buckets be too large, the tennis balls will not be adequately separated to detect
individual frequencies. This would lead to always using the highest resolution for
spectral or band analysis. Higher resolutions require greater amounts of time to display
the spectrum, thus a balance must be reached between the capture time and the spectral
resolution.
Frequency Domain Checklist
- Overall Amplitude
- Time Base Waveform
- Orbits
|
