Background
One of the more difficult areas to apply vibration successfully is to slowly rotating bearings. A misunderstanding of the basics such as sensor selection, measurement procedure and the FFT process itself has often led people to proclaim that measurements on machinery rotating below 60rpm is not possible. This is untrue, it is perfectly possible - as long as you understand how.
The first mistake most people make is that they use the wrong accelerometer. Consider this: at 3,000rpm, a velocity of 1mm/s corresponds to an acceleration of 0.03g, whereas at 30rpm that same velocity is now 0.0003g. If I were using my normal 100mV/g accelerometer, that would give me a voltage of just 30µV of signal in the presence of typically 20µV of noise. That is clearly not going to be easy to deal with.
The answer, of course, is to increase the sensitivity of the accelerometer to say 500mV/g. Some care is required here, and it is highly recommended that you test it before you but it because some of these accelerometers might actually be 100mV/g accelerometers with built-in 5x gain amplifiers. This is also not desirable since, in all probability, the net result of amplifying the signal and the noise is likely to be a reduced signal-to-noise instead of the increase we need.
There are also considerations in terms of electronic noise as well as the noise created by the FFT process itself. With such low frequencies, electronic noise such as Schott or 1/f (one-over-f) becomes critical. As the name suggests, this is a noise source which increases with decreasing frequency. But this noise source, which will be in my measurement chain, is in acceleration. If I integrate this signal to velocity, my 1/f becomes 1/f2 . An increase by a factor of 10 in my acceleration, for example, now becomes an increase by a factor of 100 in velocity. So integration is not a good idea.
The FFT process itself is also a source of noise at these frequencies. To overcome that noise, I must spread out the overall amount of noise present in my system across as many lines of resolution as I can, so that the noise/line is minimized.
At this stage, there is also a temptation to add averaging in order to "smooth out" even more of the noise sources. However, always remember that averaging not only reduces the effects of random noise, but it also acts as a filter for any short duration events which may result in valuable data being lost.
Increasing the resolution means increasing the amount of time it takes per reading. A 10Hz frequency range with 3,200 lines of resolution represents a block of time data 320 seconds long. If you were to take 4 averages, that is over 21 minutes per measurement!
The Ultrasound Alternative
Ultrasound is recognized as being a much quicker way of "listening" to bearing condition. Because it is working at much higher frequencies than vibration, there is no need for an ultrasound system to take the slow rotational speed into such a critical account.
The SDT170 ultrasound system used for this work measures at 38.4kHz and processes the signal to provide an audio output 2kHz wide which contains data from roughly 36-38kHz and 39-41kHz (Figure 1).
Therefore, this audio signal contains all the activity which is occurring at that much higher, inaudible, frequency. When dealing with bearing defects, what is actually going on up there is normally the broadband noise of friction caused by poor lubrication, and is impulsive or impacting in nature. In the case of bearing defects, this signal is rarely periodic, which makes them all the more difficult to detect when performing long, slow, averages.
Since the ultrasound system is listening to these impacts, the need to work using slow signal processing methods is not required.
By taking the audio output and feeding it into a vibration data collector, the SDT170 can be effectively used as an intelligent sensor. By fixing the amplitude setting of the SDT170, the measurement chain will provide repeatable and comparable data.
In fact, the only consideration is that of how much data should be gathered in order to be considered representative. For a machine rotating at 12 rpm, for example, I would want to have a sample of perhaps 30 seconds in order to be sure that I have representative data.
If I am working with an ultrasound detector in conjunction with a vibration data collector, there is also a limit caused by the selection of frequency range and sample length. Since my ultrasound system produces an output of just 2kHz, there is little point in working with any higher maximum frequency. If I work with the maximum number of lines, say 12,800, then the maximum time sample that I could acquire would be limited to 6.4 seconds.
Clearly, I now have a potential conflict of interest between what I would like to record and what I can record. I can overcome this conflict to some extent by monitoring the signal before I store it to ensure that what I have is representative. Alternatively, I can scrap the idea of using a vibration data collector as my recording device and consider instead using a digital recorder working with a .wav file.
There are devices now available in the professional audio industry which are perfect for this application - small, battery operated, manual gain, .wav file format recording to SD card - which are capable of recording 360 minutes of data onto a 4Gb memory card.
The selection of a suitable device is critical. Because of the transient nature of the signal, auto gain control and MP3 recording are going to cause problems. The automatic gain control will be constantly changing the recording level which makes it very difficult to record comparative signals. MP3 compresses the signal, elevating the level in the quieter parts and compressing the peaks. This will clearly cause corruption of the signal.
The Results
Figures 2, 3 and 4 show results taken on some crushing rolls. This type of machine has two rolls crushing stone down into a powder. This data was taken directly from an SDT170 ultrasound system into an Adash VA3 vibration data collector. The length of the signal was 1 second. These comparison time signals compare the four bearings on one machine.
The two inboard bearings in Figure 2 are of particular interest - the inboard bearing on the moving roll (top left) is exhibiting a classical suspect bearing pattern where the bearing on the fixed roll (bottom right) shows a combination of suspect bearing and what was eventually defined as pocket noise - a fairly periodic noise corresponding to 84x rotational speed.
Figure 3 shows a suspect bearing on the inboard bearing on the fixed roll (bottom right) - note the strong impulsive nature of the signal and the lack of any periodic nature.
There is often an interesting, and sometimes heated, debate on the topic of whether measurements should be taken in loaded or unloaded conditions in this application. The argument for unloaded is normally a concern about the random nature of the operating process, which makes it easy to hide a defect in the presence of all the noise.
The counter position, which I support, is that only in real-life operating conditions do we see the true forces being experienced by the machine. The comparison chart in Figure 4 compares the same two bearings with measurements in loaded and unloaded conditions.
This difference is clearly enormous. Measurement in the unloaded condition would never convey the nature of the extremely high forces present in normal operation - surely providing an unfortunate false sense of security.
Conclusions
A portable ultrasound inspection of slowly rotating machinery can be successfully used to inspect bearing condition. The nature of the measurement method means that the time taken to survey such bearings is significantly reduced.
Analyzing the audio output signal for an ultrasound data collector provides diagnostic information in a fraction of the time required to collect and analyze similar information from a vibration measurement.
Tom Murphy is an Acoustics graduate from Salford University and has 25 years experience in the world of industrial ultrasound and vibration measurement - 15 of those years have been involved with the use of Operating Deflection Shape techniques in the paper, printing, petrochemical, power generation, pharmaceutical and food industries. Tom is the Managing Director of Adash 3TP Limited, based in Manchester, England, a Company specializing in the application of vibration, infrared and ultrasonic technologies to improve maintenance. More info can be found at www.reliabilityteam.com and Tom can be contacted at +044 161 788 9927
