The field of vibration analysis was first developed in an era that had very little of the technological capabilities that we have today. As instrumentation advanced from the reed vibrometer, to the tunable vibration meter, to the swept filter analyzer, to the FFT, to the first real time analyzers and finally to the multichannel data collectors and analyzers we have today, at each stage the existing technology limited the advancements in vibration analysis. As technology advanced, it opened the doors for new applications which led to better analysis. With recent advances in computer and sensor technology some of the practical limitations of the past have become obsolete. However, some of the paradigms have not changed along with the technology. Taking advantage of the new capabilities can move vibration analysis forward in the development of the field.
Most of these technological advances were not developed primarily for the relatively small vibration market. Microchip based sensors, for example, including accelerometers, have many applications outside our field. Millions of mobile phones using such chip based accelerometers have driven companies to spend vast amounts of resources to develop better chips in a wider variety of capabilities. Wireless technology would be a decade away if it was only for collecting vibration data, but today for an instrument OEM to add a small Bluetooth module to a device is simple and inexpensive. Other developments are simply the ongoing advancements in computer technology, with exponential growth in processing speed and storage.
MEMS Accelerometers – an Alternative to Piezoelectric
With the development of micro electrical mechanical sensors, MEMS, new possibilities now exist for taking vibration measurements. Robust, low cost and very small, MEMS accelerometers can be mounted in many places where traditional transducers would not fit. They have relatively low noise and a flat noise over their useable range. They are not sensitive to heat and are quit shock resistant. Because MEMS sensors do not need a mass, the settling time after attaching the transducer with a magnet is nanoseconds. This simplifies the data collecting process and reduces the time between each measurement. Due to faster A/D converters and increased processor speeds, multi-channel instruments can be run on tablet PCs.
Fig. 1 Triaxial MEMS accelerometer
Courtesy of Update International, Inc.
Obsolete: The Need for Averaging
In the past, a standard vibration measurement for periodic monitoring would be taken with 400 lines of resolution and 8 averages. This has changed to 800 or 1600 lines and 4 averages. The averaging is to reduce the affects of transients that may be in the time signal. The reason the number of averages recommended is reduced from 8 to 4 as the number of lines is increased from 400 to 800 is so that a similar amount of total averaging is applied.
When a vibration signal is processed by the FFT, the resulting amplitude of each frequency bin is the average amplitude of that frequency over the period of time the vibration was taken. If a transient vibration occurs during a short time period, the average amplitude is still significant at that frequency. If that spectrum is averaged with spectra from periods where that transient did not occur, then the resulting amplitude at the transient is diminished while the amplitudes at periodic, relatively constant frequencies remain essentially the same. The result is a short period, well averaged spectrum with low resolution.
If the vibration is taken at higher resolution (using more lines of resolution) which requires more seconds of time signal, then in each bin the amplitudes are the average over a longer time period. Any transient which occurs during that period is reduced as the amplitudes are the average over a longer period. Therefore, less averaging of the resulting spectra is necessary to diminish transients. The final result is a well averaged spectrum with higher resolution. Table 1 shows a comparison of three measurements with different resolutions and averages.
| F max 120 kcpm; 50% overlap | Time of each period measured | Total time measured | Resolution per bin |
| 400 lines, 8 averages | 0.2 seconds | 0.9 seconds | 5 Hz, 300 cpm |
| 1600 lines, 4 averages | 0.8 seconds | 2.0 seconds | 1,25 Hz, 75 cpm |
| 12,800 Lines, 1 average (no averaging) | 6.4 seconds | 6.4 seconds | 0.15 Hz, 9.375 cpm |
| F max 180 kcpm; 50% overlap | Time of each period measured | Total time measured | Resolution per bin |
| 400 lines, 8 averages | 0.13 seconds | 0.6 seconds | 7.5 Hz, 450 cpm |
| 1600 lines, 4 averages | 0.53 seconds | 1.33 seconds | 1.875 Hz, 112.5 cpm |
| 12,800 Lines, 1 average | 4.27 seconds | 4.27 seconds | 0.23 Hz, 14 cpm |
Table 1
Obsolete: The Expediency of Separate Waveform and Spectrum Files
Traditionally, files are saved as waveform files or spectrum files. This allowed smaller file sizes and faster recall if, for example, one wanted only the FFT plot. Typically spectra and waveforms used different time periods and sampling frequencies. The primary reason for limiting TWF measurements in the past was to keep file sizes small, which is no longer necessary, as will be covered in the next topic. Vibration measurements can also be saved as raw, digitized waveform data. Parameters can be set to meet the requirements for viewing the data in both the frequency and time domains. The software can process the waveform data through the FFT when necessary in just a few seconds. This allows the analyst to switch between viewing the vibration as a waveform, a spectrum and even listening to it as a sound file all from one data file. This will also allow post processing the data in new ways, such as applying different filters or windows long after the data has been captured and stored.
Obsolete: Lower Resolution for Smaller File Sizes
The file size of a higher resolution vibration measurement will be larger than one of lower resolution (regardless of averaging). Keeping the file sizes small was considered necessity in the past; however this should no longer have to be a concern. Computer storage density has increased at such exponential rates that any system taking advantage of the modern technology would be hard pressed to fill up the storage that is inexpensively available (see table 2).
Measurements with an Fmax of 600,000 cpm; 12 measurements per unit train taken once a month on 500 unit trains:
| Seconds of waveform | Resolution CPM/bin | File size in KBs | GBs of storage required / year |
| 1 | 60 | 158 | 12 |
| 3 | 20 | 471 | 34 |
| 6 | 10 | 940 | 68 |
| 12 | 5 | 1878 | 136 |
| 30 | 2 | 4696 | 338 |
| 60 | 1 | 9391 | 676 |
(1 terabyte = 1,000 gigabytes)
Table 2
As of this writing flash memory USB drives of 512 gigabytes are sold for under $20 and external hard drives with 1 Terabyte for under $55. By the time one fills that much storage with route data, the available capacities will be greater and the prices lower.
It is impractical to expect a specialty instruments such as data collectors which are made in relatively small volumes to constantly be available in the field without their using “yesterday’s technology” such as not-quite-state-of-the-art processors. However, the capabilities of modern instruments are still way beyond what was around when vibration data collectors were first being developed. At that time the use of the FFT (Fast Fourier Transform) algorithm was needed to increase the speed of the processing of digitized vibration data using computer processors that were slow by today’s standards. Today the processors are so fast that the relatively small increase in calculation speed provided by using the FFT algorithm is unnecessary. By using the DTF (Discrete Fourier Transform) any number of lines of resolution can be used, giving more flexibility to the user. One enters the desired resolution or the number of seconds of waveform data to take for a given Fmax and the software could determine and apply the number of lines of resolution required. Any number of lines could be used, not just those that meet the FFT criteria.
How Higher Resolution can Facilitate more Accurate Analysis
The following example show how better resolution can result in more accurate analysis without bogging down the data collection process. An analyst reviews vibration taken on a motor and pump. The 1200 nominal RPM motor is connected with a 3 jaw coupling to a fan with 6 blades. The spectrum in figure 2 has an Fmax of 120,000 cpm and 800 lines of resolution. Several harmonics of running speed are present.
Figure 2
Nearby on the same deck is an 1800 RPM motor coupled to a blower. Some of the vibration from that machine train reaches the motor and fan. Table 3 compares the exact RPMs and harmonics of both machine trains.
| Motor/fan |
| 1x 1187 |
| 2x 2374 |
| 3x 3561 |
| 4x 4748 |
| 5x 5935 |
| 6x 7122 |
| 7x 8309 |
| 8x 9496 |
| Motor/Blower |
| 1x 1783 |
| 2x 3566 |
| 3x 5349 |
| 4x 7132 |
| 5x 8915 |
Table 3
With a resolution of 150 cpm per bin, it’s not possible to know if the fourth peak is 3 times the fan RPM or 2 times the blower, nor identify the peak that appears to be the 6times the fan or 4 times the blower.
Increasing the lines of resolution to 6400 lines as seen in figure 3, reveals that the peak near 6x fan is actually three frequencies. The largest peak is 6 times the fan speed which is the blade pass frequency. There is a small peak at 4 times blower speed and another at 7200 cpm which is 2 times line frequency.
Figure 3
Even at 6400 lines there is still not high enough resolution to identify the peak near 3 times fan speed. Increasing the number of lines to 12,800 as seen in figure 4, reveals that there is a smaller peak at 3 times fan which could be from the 3 jaw coupling and another larger peak at 2 times blower speed, which could be from misalignment between the motor and blower. This resolution would require only 6.4 seconds of waveform data.
Figure 4
Other situations can cause vibration frequencies that are very close to each other. For example, some bearing fault frequencies are very close to a harmonic of the shaft RPM, though they should never be exactly synchronous. Sidebands can coincidently be close to other unrelated frequencies. An analyst is not always alerted to the possibility of more than one frequency being present in one spectral bin. Taking higher resolution data for both monitoring and detailed investigations will give the analyst a more accurate picture of the vibration and that is an important advantage.
Conclusion
With the advances in technology that have occurred in recent years many of the practical restrictions on data collecting that have existed since the first data collectors were put into use have now become obsolete. Processing speed and storage capabilities have increased the ability to obtain and store data with higher resolution than ever before. Saving raw waveform files gives the analyst more options and flexibility with post processing the data. Faster processors make the FFT unnecessary. Taking advantage of these developments does not require cutting edge technology that is expensive, hard to use and typically used only by scientists and researchers. The hardware is available in common tablet computers with off the shelf sensors. Creative thinking and innovation can make use of those resources. It’s time to take advantage of those possibilities.