|
Case
study
Monitoring
of Slow Speed Bearings
using
the Microlog CMVA60 ULS (Ultra Low Speed)
by
Dr Bob Jones, SKF Condition Monitoring
Abstract
With
the advent of Enveloped Acceleration, it is possible to monitor the
operation of slow speed equipment and its rolling element bearings. For
purposes of this paper, we will consider slow speed to be anything
rotating less than 100 RPM. The author has collected and analyzed the
condition of bearings in a range from 0.1 RPM to 8.3 RPM with satisfactory
results. This involves the use of the time domain as well as the frequency
domain. This paper will discuss techniques for setting up the data
collection and give examples collected in the field. With the release of
the SKF Microlog CMVA 60 ULS (Ultra Low Speed), slow speed equipment may
be monitored, in velocity, down to 6 RPM. Examples will also be shown
using this equipment.
Introduction
Monitoring
of slow speed equipment bearings is a mixture of engineering and art. Some
would say it's magic and just let it go at that, but in fact it is a
finely tuned engineering process that relies on some of the latest
electronic processing technology. This technology allows us to enhance the
extremely low level signals generated by the bearing rotating elements as
they make contact with a damaged area of the bearing ring or cage. In the
past we used the measurement of velocity to attempt to analyze the
condition of slow speed bearings. Sometimes we were successful and
sometimes we weren't. Everyone in this business can probably recall a
horror story where after carefully monitoring a bearing over a period of
several weeks and declaring it to be in acceptable condition, have it fail
shortly thereafter. The weakness of using a velocity measurement for
bearings is in the mathematics.
Mathematically
velocity is a product of dividing distance by time, V = d/t. This gives
you inches per second, millimeters per second or miles per hour, whatever
units you wish to use. In any rolling element bearings, the "d"
is a very small number, in the order of 0.0005 inches or less. If the time
element is relatively large, 0.5 seconds for example. This would result in
an element velocity of 0.001 IPS. In large machinery as in paper mills,
screw conveyors, or large slew bearings, the bearing elements do not have
a large mass relative to the machine. The velocity generated has to be
transferred to the bearing housing and it's like tapping on it with a tiny
hammer, the bearing housing movement is even smaller and the vibration
measurement probe just doesn't detect it. In laboratory work we have taken
velocity measurements on severely damaged bearings that would have to be
replaced in an operational environment. The readings we obtained on these
damaged bearings was in the order of 0.0027 IPS, at 1770 RPM! Certainly
not an amplitude that would cause concern and yet we knew the bearings
needed replacement.
We
can see then that at the site of the damage, the elements just don't have
enough distance to develop any velocity (speed), or mass to transfer an
energy pulse that is consistently measurable with any portable data
collector. This doesn't say that one can never see a bearing problem using
velocity, but our experience has been that when you can see it in
velocity, you are already in trouble, and probably have been for awhile,
you just didn't know it. This means that we need another method to measure
what is happening inside the bearing while it is in service. Using
proprietary algorithms, SKF Condition Monitoring has produced in the CMVA
55 and the CMVA 60 Microlog data collectors, instruments that have the
ability, by using enveloped acceleration processing, to detect initial
damage in rolling element bearings that are of a size that literally
cannot be seen without magnification.
Enveloped
Acceleration (gE)
Before
we examine case histories of slow speed equipment, it is important that we
understand why and how enveloped acceleration is a superior method for
analyzing bearings versus using velocity measurements. We will do this
without delving into the specific mathematical algorithms and electronic
methodology that is used in the process.
If
we recall our high school physics, we remember that F = MA, force is equal
to mass times acceleration. Of these three units, we can measure, using an
accelerometer, the acceleration. If we arbitrarily set the mass to equal
one, then the force is equal to the acceleration and we now have a method
to measure the forces occurring inside the bearing. Therefore, what we
will be measuring is the repetitive forces generated as a rolling element
impacts a flaw or damaged site inside the bearing. If we attempt to
measure this force with standard acceleration (Gs), we will be attempting
to capture a single event of such small amplitude that, as with velocity,
may not be observable.
In
a nut shell, enveloping allows us to enhance the repetitive signal
produced as a bearing element passes over a damaged area in a bearing and
degrade the non-repetitive signals. This is accomplished by using one of
four specific filters to capture the harmonic energy in the range of that
filter. After this capture, the energy is electronically processed to
provide the enhanced signal by mathematically summing these harmonic
signals. Since the user has set up a specific frequency range to examine,
this captured energy is then demodulated and presented to the user in the
frequency range selected. Please note that there is no connection between
the specific filter range selected prior to the data capture and the
frequency range selected for the displayed spectrum.
Figures
1 - 3 illustrate the four filters and how the flaw generated energy
"moves" around and may be detected better in one filter today
and another next month. It may be in all four filters, but probably one
will show it better. These examples do not show the overlap between
filters i.e., the first filter ends at 100 Hz and the second filter begins
at 50 Hz., but are satisfactory for this depiction. The data have since
been lost, but we once examined a 27 RPM machine where we had a good
bearing defect signal in each of the four filters. The choice is yours.
In
the case shown in Figure 2, the best choice would be Filter 3 since that
is the filter that would capture the most energy and produce the best
spectrum.
The
example in Figure 3 would be a case where you could use any of the first
three filters and even the fourth filter will produce a spectrum. This
could be an example of the machine that was rotating at 27 RPM and
produced acceptable bearing flaw signals using any of the four filters.
It
is mandatory in collecting vibration signals from slow speed equipment
that the user save the time domain spectrum. The data collector has
already produced one in order to convert it into the frequency domain
using a Fast Fourier Transform (FFT). The user only has to setup the point
to save the information. In one example we will look at later, the bearing
is rotating at 0.5 RPM. The bearing race defect frequency was calculated
to be 6.9 CPM which means that a roller passed over the flaw every 12
seconds or 5 impacts for the 60 seconds it took to collect the 500 CPM
spectrum. In the frequency spectrum it is not possible to see the defect
frequency and due to time constraints we could not decrease the Fmax or
increase the number of lines for improved resolution. The solution is in
the time domain, as will be shown.
In
the early development and testing of the enveloping process, the criteria
for selecting which of the four filters to use was based on the rotating
speed of the machine. These criteria are still valid for machines
operating above 500 RPM but experience has shown us that when analyzing
slow speed equipment, the rules have to be modified. For example with the
0.5 RPM machine, using early guidelines, we would have selected the first
filter with a range of 5-100 Hz. If we had only used that one filter we
probably would have missed the call. The filter that displayed the best
results was the third filter, 500-10K Hz. However, on other similar
occasions, it has been seen that the first filter was the best to use. And
recently with equipment at 2.4 RPM, the best filter to use was the second.
This experience leads to the conclusion that with slow speed equipment it
is necessary to collect three readings for each bearing, each reading
using a different filter. After it is determined which filter is providing
the best spectrum, the data collection setup should be used and the other
two disabled. However, if over time the signal changes with either an
increase or a decrease in amplitude, the disabled points should be enabled
and another evaluation conducted.
Laboratory
testing disclosed the reason for this shifting. Early on it was assumed
that slower speeds would generate lower frequency signals and as the speed
increased, the energy would move upward to the higher frequencies. This
has proven not to be the case in all situations. At 0.5 RPM, the energy
was clearer in the 500-10K Hz range (third filter) than it was in the
5-100 Hz range (first filter). However, that was confirmed only for that
day and that damage site. Further damage may cause the clearest energy to
move to a lower or higher frequency. Therefore, for plant personnel it is
imperative that they plan on multiple readings for critical slow speed
bearing.
Figure
4 is a typical setup screen for collecting enveloped acceleration
readings. The important points to note are the filter selection, FFT and
Time selected, one average and the start frequency set at "0".
It is possible to set the start frequency at "0" because you are
using an acceleration signal and there is no integration to produce the
ski slope that is seen and influences the overall amplitude in velocity.
In
addition to learning how to look for the energy, the user needs to
understand some of the characteristics of an enveloped spectrum. We are
all familiar with the widely published alarm limits for velocity
measurements, alarm 1 is generally 0.25 IPS and alarm 2 is 0.35 IPS. And
importantly, these are not speed dependent. In other words, these alarms
apply to a machine operating at 600 RPM as well as the machine operating
at 3600 RPM. It is mandatory that the user understand that this is not the
case with enveloped measurements. The rolling elements inside a machine
rotating at 100 RPM are going to generate significantly less energy,
because less forces are present, than the elements from a machine rotating
at 3600 RPM. Therefore, the user is required to associate the speed with
the observed amplitude.
This
point is illustrated in Figure 5. Again in the laboratory, using a DC
motor, a flaw was induced in the motor bearing. With the speed set at 50
RPM, the amplitude of the enveloped reading was 0.004 gE. As the speed was
increased, the amplitudes can be seen to increase to the point where at
3600 RPM, the amplitude had increased to 1.8 gE. That's an increase from
low to high speed of 450 times, all from the same same flaw. Clearly the
user has to consider the machine speed and be careful not to apply limits
to Machine A that were developed for Machine B that operates at a
different speed.
At
this point we have selected a bearing to monitor, built three points using
each of the three filters with the time waveform saved, assigned the
bearing to the point and are ready to go collect some data. Once you have
the data, here is what to look for. From the bearing defect frequency data
base, you should be aware of what frequencies you are looking for.
But,
in many cases you will not be able to know for certain what bearing is
installed because the machine has been overhauled numerous times and no
records have been kept. On the basis that a good engineering guess is
better than no information at all, do the following: Look in the frequency
spectrum for harmonic signals. Harmonic content in the frequency spectrum
tells us that the energy from the time wave is being clipped or truncated.
A Fourier transform of a clipped sine wave will produce harmonies. If
there is no damage, there is no clipping and no harmonics. Therefore if we
have harmonic energy that is not associated with such things as rotation
speed, gear mesh or twice line frequency, we are probably seeing energy
from a damaged bearing.
The
Envelope Process, Non-Technically
We've
discussed the filters and the energy they capture. Now we will examine how
this energy is processed to provide us the spectrum used for evaluation.
Although the time domain (Figure 6) has time as the "X" axis,
the algorithms enable the data collector to select the desired filter
frequency range data from this time information. The Fmax was set to 80 Hz
in order to collect 5 seconds of data. If you need more time, the Fmax is
reduced. If less time is needed, the Fmax is increased.
Figure
7 is a representation of what is happening inside the data collector in
processing this energy into the enveloped spectrum based on using the
third filter which has a range of 500 Hz to 10,000 Hz.
The
filter has now assembled, out of the collected data, the harmonic content
of the energy being emitted from the bearing. In the Figure 7 example, the
bearing has a fundamental frequency at 120 Hz. The harmonics then follow
out to the limit of the accelerometer although we cut it off at 15 kHz.
The filter has captured 80 harmonics which are then mathematically
processed to produce an enveloped spectrum similar to the next example
which is from a damaged paper roll bearing turning at 88.5 RPM.
To
continue the process with another spectrum, we see in Figure 8 what
appears to be multiple harmonics. Using "Set Speed", mark the
rotating speed of the shaft exactly as possible, this is 1X. Then set the
single cursor on the first harmonic and observe how many orders of running
speed are shown in the "Single Value" window. If the orders
shown are a non-integer number i.e., 6.532 or 4.575 orders, you probably
have a damaged bearing. The reason for this is in the mathematics of
calculating bearing defect frequencies. It is not true in all cases
because some defect frequencies will be seen to be 4.012 or 5.991, so
close to an integer that this evaluation is not valid. It is also not
valid if you are working with a gearbox with multiple internal shafts.
However, we have called bad bearings based solely on the order information
when the owner did not know what bearing was installed.
Figure
9 is a spectrum from a recently rebuilt 500 HP with a variable frequency
drive motor turning a screw mixer at 674 RPM. The frequency spectrum, with
harmonics, and the first harmonic at 10.6825 orders of shaft speed, led to
the call of a bad bearing although at the time the owner did not know what
type of bearing was installed. The motor shop, using velocity measurements
didn't agree, but the owner insisted. When the bearing was pulled, the
damage was found. Apparently on installation, the installer had slipped
and damaged the lip of the inner ring. As the rolling elements passed over
the area, the inner race defect frequency (BPFI) was generated. Score one
for the good guys. Now some of you sharp eyed readers are going to notice
that the CPM in the Single Value window is 7200 CPM and will say that's
twice line frequency and indicates an electrical problem in the motor
stator. There's another thing you check for in cases like this. When you
use the "Set Speed" marker, place it on the first harmonic and
check the exact frequency. Remember the single cursor goes from line to
line and this frequency is between lines. "Set Speed" check
showed it to actually to be 7211 CPM, too high to be an electrically
induced problem. When we examine some slow speed bearings you'll see
similar examples using the orders as a clue.
There
is a characteristic of enveloping that will often give us a problem when
we are trying to use the "Set Speed" and mark the 1X on an
enveloped spectrum. The enveloper is looking for harmonic content. If the
unit is well balanced and there is no looseness, the rotating forces will
not generate any 1X forces and will not display a clear peak at the
rotation speed. Even if the unit is unbalanced, the once per rotation out
of balance forces do not generate harmonics. So the question is: How do we
know where to mark the rotation speed on an enveloped spectrum? The
solution is to also collect a high resolution velocity spectrum
immediately after collecting the enveloped spectrum. All rotating
equipment has some out of balance forces which can be seen in the velocity
spectrum. Put the cursor on this 1X signal, use "Set Speed" and
determine the exact speed. Return to the enveloped spectrum and again open
the "Set Speed" window. Type in the exact speed from the
velocity spectrum, hit the green check mark and this speed will be marked
on the enveloped spectrum at 1X, irregardless of the position of the
cursor.
Some
examples of this technique, are shown in Figures 10 and 11.
Taking
the original spectrum, we have added in Figure 12 the harmonic marker, the
rotation speed as determined from the velocity spectrum, and displayed the
"Set Speed" window with the cursor on the suspect frequency.
Here you can see what we discussed earlier, the cursor is on the line at
7200 but in actuality is at 7210.6 CPM. And, the "Single Value"
window shows the orders at 10.6825. In this case since the motor had been
rebuilt recently, we did know the bearings that were installed so to seal
the inspection, the next spectrum is with the bearing frequencies (BPFO)
overlaid. One point needs to be noted here. This was a fairly new bearing
and so does not have any significant wear. Therefore, the bearing
frequency overlays match up very well. This may not be the case when you
are testing old, worn bearings. The Frequency Analysis Module (FAM)
contains frequencies for new, perfect bearings. If the spectrum is from a
worn bearing, it is probable that they will not match up as in this
example. The answer is to also use the other information you have to
arrive at a conclusion, don't use just the overlays.
Slow
Speed Applications
A
major automobile parts plant produces diesel engines. The production line
is moved by four very large gear drives with an output shaft speed of 8.3
to 12.0 RPM. The output shaft bearings were a constant maintenance problem
in that they were internal and on the bottom of the gear case. Although
the plant has an excellent vibration monitoring program with numerous
"saves" on other equipment, this particular gearbox would fail
without warning, causing a three day unscheduled outage for replacement.
The automotive business operates on a "just-in-time" assembly
basis so they were required to maintain in stock a three day supply of
engines to ship while the repairs were made. We demonstrated to them with
the following spectra just what enveloped acceleration could do for them
in monitoring slow speed bearings. As part of the demonstration we
collected three readings, acceleration, velocity, and enveloped
acceleration. The spectra speak for themselves Figure 13.
We
overlaid the inner race defect to show that although there may be a slight
amount of energy at these frequencies, not many people are going to push
for action with an amplitude of 0.00005 G's!!!
The
velocity spectrum Figure 14 shows us nothing of concern. And, this was
their problem, they had a large amount of velocity data that didn't show
them anything of value.
As
will be seen in the enveloped spectrum Figure 15, there is damage in this
bearing but there is nothing of concern showing in the velocity
measurement. There was a speed change to 11.1 RPM but as can be clearly
seen, there is harmonic activity and the FAM overlay aligns very well to
show us that there is damage in the inner race. One of the first comments
made was that the amplitudes are too low to worry about. This is not true.
Return to Figure 5 and note that this amplitude 0.006 at 11 RPM would be
probably be equal to 1.5 to 2.0 eG at 3600 RPM, certainly a cause for
alarm. We knew the bearing that was installed, but our first clue was
while collecting the data we saw the apparent harmonic signals.
This
large gearbox had other bearings of course and for comparison purposes, we
collected data from another bearing, which based on the harmonic content,
appeared to have damage. With the application of the bearing frequency
overlay, it appeared there was in fact damage on the outer race Figure 16.
A
follow up report told us that they had found both bearings to be damaged
as detected with the enveloped acceleration spectra Figure 17.
In
all cases, the two most important items to look for is the harmonic
content and the non-integer order of rotation speed. Then if the bearing
information is available, use the overlay function, FAM, to confirm your
data.
Slow
Speed Analysis Using Time Domain
With
very slow rotating equipment such as in paper and steel mills, large slew
bearings and cement kilns, there may not be enough energy generated to
produce a signal in the frequency domain. In these cases it is possible to
detect bearing flaws in the time domain.
What
is necessary is to collect a spectrum of long enough duration so that the
rolling elements pass over the flaw five or six times. Sometimes the
mechanics of the machine will not allow a full 360 degree rotation such as
a radar azimuth bearing and in those cases, it may not be possible to
collect enough data for a frequency or time analysis. In those cases it is
necessary to use amplitude trending. The Fmax is set so that the time span
in the time domain is large enough so that data is collected while the
unit is moving and ends prior to the machine stopping.
In
a small steel mill, trending of the overall amplitude in acceleration had
resulted in a steady increase in the slope of the curve, but there was no
indication of specific damage. The bearing frequency for the outer race
was calculated to be 6.9 CPM at the 0.5 RPM rotation speed. Using
enveloped acceleration the following time domain was collected Figure
18.
What
we are interested in are the bursts of energy that appear in the spectrum.
By using the harmonic markers and placing each one approximately in the
middle of a burst of energy, the time interval between bursts is measured
and since frequency is a reciprocal of time, it also provides a
frequency. In this case, the frequency is determined to be 6.8 CPM, close
enough to the calculated frequency to call it as the flaw and source of
the detected energy. What is occurring is that as a rolling element passes
over a flaw, the burst of energy is created and the time between bursts
gives us the frequency.
The
question may arise concerning the lack of signal for the first 12 seconds
or so. When the bearing was removed it was found to be distorted because
the expansion assembly had locked and the heat expansion had to go
somewhere. It distorted the bearing so that the rollers didn't always make
contact with the race. When they did we got the burst of energy, when they
didn't, no signal. We checked the signal using all three filters and this
data was collected using the third filter. The same signal in the
frequency domain produced this spectrum Figure 19.
The
alternative to doing this of course is to set the Fmax to about 100 CPM
but this would have required 4 minutes of data collection and we couldn't
rotate the vessel for that long. For comparison the final two spectra are
from a similar vessel with new bearings Figures 20 and 21.
Conclusion
Enveloping
of the acceleration signal is the best method available today to analyze
bearings. It does require care in the application of the procedures, as
with any technique using computer technology.
Top
of page |
Back
to Case Studies
|
