Frequency analysis is an important method of vibration analysis. It provides information on vibration sources and helps identify those components in the vibration signal that are often small, but nevertheless important, for diagnosis. Each vibration can be attributed to a particular excitation source or machine part.

Irregularities or damage lead to changes in the frequency spectra's appearance. These faults include mechanical or thermal unbalances, characteristic vibrations, changes in alignment, defective bearings, gear tooth damage, changes in play, frictional corrosion, tears, etc.

Digital FFT analyzers are state of the art. Fast Fourier Transformation (FFT) is a mathematical procedure that automatically occurs within the measuring device and produces frequency spectra.

FFT analyzers display spectra as line graphs. To achieve a high frequency resolution, the number of lines (LOR) should be as high as possible. While the minimum number of lines required is 800, it is preferable to have more. Of course, time to collect the spectra will need to be considered since the higher amount of time to collect data is equal to the higher number of lines that data is collected. For most standard operating machines, 1,600 LOR is normally used for adequate data collection and for faster data collection.

The amplitudes of the individual frequencies are displayed during measurement. They can be plotted on a linear or logarithmic scale for each component of the spectrum. The linear display shows magnitude only. The logarithmic display, on the other hand, shows even the smallest vibrations. The prerequisite is that the frequency analyzer have an amplitude resolution of at least 100 dB.

Likewise, the frequency scale can also be either linear or logarithmic. It should be noted that it is virtually impossible to compare frequency diagrams plotted on a different scale even if they contain the same information. We recommend a linear scale for displaying amplitude and frequency.

  Different representations of frequency spectra

 

Different representations of frequency spectra

How to Measure Valid Frequency Spectra

When dealing with FFT spectra, there is a danger of incorrect measurements that can result in diagnostic errors. In fact, almost all suppliers of frequency analyzers offer special training courses and thick manuals. This still does not prevent false measurements from occurring frequently-particularly in the case of gear analyses-as a result of insufficient measurement technology. Pay attention to the following settings while taking measurements.

Adequate Measurement Periods

An effective measurement period is decisive in the analysis of vibration signals. If loads change, e.g. as in wind powered systems as a result of wind gusts, measurement periods of up to five minutes may be required for reproducible measurements. For constant speed rotors and RPMs greater than 200 rpm, we recommend an effective measurement period of at least 10 seconds and, at RPMs greater than 20 rpm, a measurement period of at least 60 seconds. A measurement period of 60 seconds at 20 rpm means 20 revolutions are the basis for the technical estimation of the rotating shaft vibration. Since few analyzers allow the measurement period for measurements within the frequency range to be adjusted directly, the number of averaging must be set high until the measurement periods given above are reached.

High Enough Frequency Ranges

The requirement for long measurement periods and high frequency ranges places the highest demands on the frequency analyzer. For example, if two channel vibrations up to 2,000 Hz are recorded for five minutes, 1 MB of measurement data is collected to be processed. Usually, the upper limits of the frequency range under consideration have to be greater than the number of revolutions by a factor of approximately 100. Alternatively, the frequency range can be broken down into several sections and carried out in several measurements, but this takes time.

Recommended Frequency Resolution

The analysis of sidebands is an important tool in machine diagnosis. For the sidebands to be adequately resolved, the number of lines must be correspondingly high. Thus, a frequency resolution of 1,600 lines is recommended for standard machines, and 8,200 lines for multiple stage high-performance gears. Alternatively, the frequency range under analysis can be dissected into several small sections again and perform several measurements with a smaller line count. However, the time required increases considerably.

Sufficiently High Amplitude Resolution

If individual frequency components are lost in the noise produced by the measurement technology, the measurement technician misses information. On cooling tower drives, for example, strong winds can cause vibrations of up to 15 mm/s, while the actual gear vibrations in the final gear drive stage are 0.015 mm/s, i.e. 60 dB weaker. Therefore, the measuring system should have a significantly higher dynamic response (e.g., 100 dB amplitude resolution). Alternatively, the user can attempt to find ranges where the interference vibrations do not occur so strongly or where the specific gear vibrations are higher by using more measurement locations, but this also increases the measurement overhead.

Correct Window Function

Harmonic and square-wave vibrations must be evaluated differently in FFT frequency analysis. Information is lost if the correct window functions for the FFT algorithm are not used. Digital filters can be better adapted to the specifics and give better results. We recommend using a Hanning filter as the window function, and not to change this as far as possible.

 

Pumps & Systems, August 2010