Slow Speed Bearing Analysis –

monitoring in the grey zone

Nick Williams For many plant processes operating in

Advanced Maintenance Solutions Ltd, today’s environment of maximum return
Newport, Gwent at minimum cost, this widely accepted
vibration analysis rule set excludes from the
monitoring programme the very items that
need to be regularly examined. So where
does this machine speed limit come from?
While accepting the general validity of the usual
Abstract

A very high proportion of today’s

recommendation regarding vibration analysis data, i.e. monitoring programmes will utilise piezo-
that it should be gathered under repeatable conditions electric industrial accelerometers as the
vibration sensors. These are relatively small,
of speed and load, and at speeds in excess of 120 RPM, cost effective, solid state, and rugged – and
the author nevertheless advocates that experimentation are easy to install as part of a new or retro-
fit installation – and most modern portable
with the application of available techniques to the health vibration analysers will be supplied with an
monitoring of variable speed or slow speed equipment accelerometer as standard. In addition, the
problems associated with the installation
may sometimes offer beneficial results, particularly when of alternative transducers for slow speed
the alternative might be no monitoring at all. He supports applications, such as eddy current probes,
or LVDTs (Linear Variable Differential
his contention with several case studies drawn from his Transformers) – which are technically
own practice. superior in slow speed applications
– require machine-specific mounting
arrangements and prepared target areas,
leaving the piezo-electric accelerometer as a
much more attractive solution on grounds of
cost, and thus the sensor of choice for most
fixed monitoring installations. However,

V
ibration analysis has rapidly
established itself as one of the most
useful and important items in the
modern mechanical engineering condition-
based maintenance tool box. The technique
is relatively easy to apply, with appropriate
equipment, and the recorded data is quickly
available for analysis on the ubiquitous
personal computer – but all too often
the technique is reserved for monitoring
ancillary equipment rather than the vital
production drives.
The perceived wisdom, passed down
to most vibration analysts during basic
training, is that all data should be gathered
under repeatable conditions of speed and
load, and at speeds in excess of 120 RPM.
This ensures that data sets gathered within
a routine monitoring programme can be
directly compared with each other, with no
fear of amplitudes being unduly affected
by process changes, and that the vibration
data gathered is unlikely to be compromised
at the lower limits of an accelerometer’s
frequency range.
This approach to condition monitoring
allows the use of amplitude trend analysis,
both for the assessment of overall
parameters and of discrete frequency bands
within the gathered data. If repeatability
in process conditions is assured it follows
that the only variable in the monitoring
process will be the machines’ health, so any
increase in levels must be due to parameter
deterioration. This is a reasonably safe
attitude to monitoring, but for these reasons
many vibration analysts do not venture into
monitoring variable speed or low speed
equipment, as it is not considered viable as
part of a vibration monitoring programme.
But what if your plant production
equipment is operating at speeds below
the recognised cut off of 120 RPM, or
operates in a variable or intermittent way?
there are difficulties when it comes to using
these devices in slow speed applications –
1. The operation of the piezo-electric
accelerometer relies on continuous
movement to excite the sensing
crystal, with the majority of the general
purpose units having a low frequency
operational detection limit of around
2 Hz or 120 CPM, before the signal
output rolls off in a decaying fashion.
2. If signal processing is required to
enable frequency analysis (Fourier
Transformation), as opposed to
pure signal amplitude monitoring,
then the signal processing itself can
produce large mathematical errors
at the very low frequency end of the
spectral range. This can lead to false
data representation, which swamps
the genuine frequency data and can
dramatically alter overall amplitude
values.

28 | May/June 2008 ME | maintenance & asset management vol 23 no 3


Slow Speed
Improving Bearing
the Reliability of aAnalysis – monitoring
Coal Fired Power Plant using ain the
It is for the above reasons that the 120
CPM limit is conventionally advised as
the lower operating limit for conventional
monitoring, following an assumption that
the analyst will need to bve able to see the
machine’ rotational frequencies within
the spectra to diagnose the most common
vibration related faults, viz. imbalance
and misalignment (1 x RPM for the shaft
monitored). But is this necessarily the case
in all circumstances, or are there other
characteristics that could be assessed
instead, making the use of industry
standard components viable, enabling
process monitoring that would otherwise be
considered prohibitively expensive?
For many plants and items of
equipment the defect which is most likely
to stop a machine in its tracks is bearing
failure. The detection and resolution of
other mechanical conditions, such as
imbalance, misalignment, mechanical
looseness, resonance, etc. is important, and
will have a direct detrimental impact on
machine life cycle, but it is a bearing failure
that is the likely fi nal consequence.
The vibration amplitudes produced by
failing rolling element defects can be very
low, even for machines operating at ‘normal’
operating speed. A defective bearing
operating at 1500 RPM will frequently
not generate sufficient energy, at the fault
frequencies to breach amplitude limits
recommended by international vibration
assessment standards, until well into fault
progression, dramatically reducing the
time interval between point of detection
to point of failure. A large percentage of
bearings carrying light loads may not trip
the alert level limits, and can fail with no
warning at all, despite the use of on-line
monitoring systems expressly intended
to detect this very problem. So if overall
amplitudes are not sensitive enough, even
at conventional operating speeds, what can
be done to detect the bearing deterioration
early enough to allow remedial action to be
scheduled correctly?
Perhaps the most powerful analytical
tool available within the modern
maintenance programme, and frequently
the most under-rated, is the data analyst.
Too much reliance is often placed on a
monitoring system’s ability to differentiate
between healthy operating symptoms
and those which indicate a fault mode
developing from infancy to failure. Rather
than high amplitudes, it is often the defects
PLANT &
MAINTENANCE
SAP grey Interface
– REWOP zone
frequency patterns, or the changes to a time
waveform that give a fault presence away.
I painfully recall being responsible, early
in my monitoring career, for monitoring an
11 KV, 11,000 HP steel mill drive gearbox,
which failed with no apparent warning of
any difficulty despite regular monitoring.
A one metre diameter output shaft bearing,
supporting a gear wheel five metres in
diameter, had disintegrated, damaging
the shaft in the process. A check of the
overall vibration trend showed no change in
amplitudes for the previous twelve months,
but when the frequency spectra recorded
were checked there in the spectra was the
tell-tale pre-defi ned bearing defect, with
a total vibration amplitude of just 0.1mm/
sec rms. Consider this in light of the closest
international standards recommended fault
limit of 7.2mm/sec rms, and the problems
with overall amplitude as the sole health
indicator start to become clear.
This was my fi rst experience with a
slow speed bearing failing catastrophically,
but my experiences over the past twenty
years of monitoring indicate that bearing
failure seldom generates high amplitudes
in low speed applications. From a positive
perspective post-failure analysis indicated
that the symptoms were present, and
accordingly the method of analysing the
frequency data was altered, in this case, to
actively seek symptoms within the recorded
frequency spectra, or time waveforms –
which simply shouldn’t be there.
In terms of failure mode symptoms, the
rolling element bearing is very obliging. It
produces a series of pre-defi nable vibration
characteristics which are directly related to
the unit’s geometry and speed of rotation.
These frequencies are readily available from
the bearing manufacturers themselves, as
they strive to demonstrate their customer
support and awareness, so the identification
of the frequency component within a
bearing vibration signature isn’t difficult,
provided the unit’s relevant speed details for
the data are known. This fault signature can
be used to accurately pinpoint a faulty unit
within an assembly cluster, and can even
be subtle enough to enable differentiation
between manufacturers of equivalent
replacements.
As a bearing fault mode deteriorates
it passes through a series of vibration
phases, and each can be helpful in fl agging,
and then tracking, a bearing’s progress
from fault inception to fi nal failure. The
vol 23 no 3 maintenance & asset management | May/June 2008 ME | 29
following description is a simplified
account of a common fatigue wear path
for a failing rolling element bearing, which
can be used by an analyst in determining
where a component is in its deterioration
progression.
Stage 1: A rolling element component
fault frequently starts as a crack
developing in the sub-surface
material. As the rolling elements pass
over the defect site, encouraging the
crack to develop, the bearing material
is excited and produces a defect
frequency, which is transmitted
through the bearing material, and
the support structure of the machine.
The action of the succession of rolling
elements travelling continuously
over the fault zone can produce a
very high number of low-amplitude
frequency harmonics, extending
high into the frequency range.
Although low in amplitude the
signal carries sufficient energy to
excite resonant frequencies within
the bearing, the support structure,
or even the monitoring sensor, and
these resonances can amplify the
signals, making them detectable. As
the impact forces contained within
a bearing fault are one of the few
sources which produce sufficient
energy to drive these high frequency
multiples, many early warning fault
detection systems seek to exploit this
symptom.
Some of these devices simply
give out a derived signal based on
the overall amplitude at the high
frequency, but as this amplitude is
reliant on resonant amplification,
and therefore structural design
and loading, the signal will only be
repeatable for a specific application.
Other systems will use signal
processing to display frequency
content or time wave-form data to
enable an analyst to further assess
the source of the fundamental.
Stage 2 (see Figure 1): As the crack
progresses, and the bearing’s
structural integrity is compromised,
the resonant amplification
frequencies will alter and migrate
the amplification zone to lower
frequencies. It is possible to track
 M1 – Machine 1 Stage 2 Brg Defect
this second stage resonant migration.
M1 Stage 2 –G4X Shaft 02 Outboard Horizontal Again, frequency analysis can allow
Trend Display
4 5
FAULT the analyst to detect the discrete fault
PK Acc in G-s

Overall Value
4

2
3 ALERT frequencies being amplified, and so
2
1 identify the source component.
0

0 30 60 90 120 150 180 210

Days: 11-July-06 To 30-Jan-07
0.12
PK Acc in G-s

0.10
0.08
Annalyze Spectrum
3-Aug-06 10:23:01
Stage 3 (see Figure 2): As the crack breaks
0.06
0.04
(SST-Corrected)
PK = .5866
the surface of the race, physical
0.02
0
LOAD = 100.0
RPM =501 (8.36 Hz)
movement is induced within the
0 1000 2000
Frequency in Hz
3000 4000 5000 bearing components as they travel
2.0
over the defect. At this point in the
Annalyze Waveform
fault development, detectable defect
Acc in G-s

1.0 30-Aug-06 10:23:01

0 PK = .5729
-1.0 PK(+/-) = 1.60/1.53 frequencies within the low frequency
CRESTF = 3.96
-2.0 spectra will appear, frequently with
0 40 80 120 160 200 240 280 320 Freq: 132.89
Time in Secs Ordr: 15.90 multiple harmonics. In the early part
Spec: .00087
of this stage it is not uncommon for
Figure 1
the fundamental source frequency to
remain undetectable.

M1 - Machine 2 Stage 3 Brg Defect

M1 Stge 3 - M2V Motor Inboard Vertical
P
0.12
Route Spectrum
Stage 4 (see Figure 3): As the fault develops
K
Ac 0.09
28-Jul-06 12:20:31
OVERALL= 1.20 V-DG
further, stress raisers at the edges of
ce
ler
PK = .2453
LOAD = 100.0
the crack allow damage and particles
ati 0.06
on
RPM = 501. (8.36 Hz)
to chip off, causing the defect to
in
G- 0.03 develop into a spall. When this is
s
0
sufficiently large, the lubrication film
0 200 400 600 800 1000 separating the bearing components
Frequency in Hz
0.8 is compromised, allowing metal to
0.6 Route Waveform
Ac
ce 0.4
28-Jul-06 12:20:31 metal contact, and frequency noise
PK = .2004
ler
ati
0.2 PK(+/-) = .5717/.5447
CRESTF= 4.03
floors within the spectra start to
on
in
0.0
-0.2
increase. It is at this point that the
G-
s -0.4 bearing’s temperature will increase
-0.6
due to the rising levels of friction
-0.8
0 100 200 300 400 500 600
Freq: 52.75
Ordr: 6.314
and the bearing, if left alone, will fail
Time in mSecs Spec: .02899
within a matter of days or weeks.

Figure 2
Of course, the fault progress path
described above is simplified, and assumes
P01 - Pump Stage 4 Brg a fatigue source rather than damage or
Pmp Stge 4-P1H Pump DE Horizontal Vel
0 .7
Route Spectrum
poor care, but given the generally accepted
PK Acceleration in G-s

0 .6 27-Sep-06 16:07:43
OVERALL = 7.17 V-DG
norms for operation, and correct design
0 .5
0 .4
PK = 2.31
LOAD = 100.0
and maintenance, this fault progression is
0 .3 RPM = 1159. (19.32 Hz) reasonably synonymous with most wear-out
0 .2 patterns for rolling element bearings and
0 .1
0
can be used to assess a bearing’s failure
0 200 400 600 800 1000 progression and enable maintenance
Frequency in Hz
20
activities to be considered based on the
15 Route Waveform
rate of deterioration through these stages.
Acceleration in G-s

27-Sep-06 16:07:43
10 PK = 6.46
5 PK(+/-) = 14.80/15.54 Given a normally loaded bearing, with no
CRESTF=3.40
0 mitigating external influences, the fault
- 5
-10
progression described can take many
-15 months, or even years to progress to the
-20 Freq: 83.13

0 50 100 150 200

Ordr: 4.303
Spec: .04212
point of failure.
Time in mSecs So, to summarise: the defect
frequency is produced by the load carrying
Figure 3
components within a bearing, i.e. the

30 | May/June 2008 ME | maintenance & asset management vol 23 no 3

Slow Speed Bearing Analysis – monitoring in the grey zone
rolling elements, travelling over the defect
– whether it is a sub-surface crack within
the race, or component damage. In short, if
there is no defect there will be no frequency
to detect, or if there is a detectable
frequency then damage is already occurring.
A further characteristic of a rolling
element defect is that it does not tend
to produce single defect frequencies in
isolation. Because of the energy involved,
and the repetitive nature of successive
rolling elements rolling over the damage,
bearing faults tend to produce multiple
harmonics, and it is frequently these
harmonics at higher frequencies that appear
before the fundamental defect source
appears within the low order spectra
A useful rule of thumb regarding
bearing defects (as a quick glance through
a manufacturer’s catalogue will tell you) is
that most defect frequencies, by virtue of
the complex geometry within the bearing
designs, are not synchronous with the
shaft carrying the bearing, unlike the
symptoms for ,say, mechanical looseness
or misalignment. Most bearing race defect
frequencies fall within the range 3 – 15
x RPM. So if a rotating element bearing
is producing a frequency with multiple
harmonics spaced at 3.15 x RPM, and
is accompanied by multiple harmonics,
provided other sources for the frequency
have been discounted, there is a very high
probability that the bearing displays a Stage
3 race fault.
Slow speed bearing components will
give out defect frequency patterns similar to
those of higher speed components, provided
the units are carrying sufficient load, and
are operating in a conventional rotational
manner. They will, however, produce much
lower amplitudes at the fault frequencies
owing to the lower levels of energy being
imparted to the bearing fault zone, and this
should be considered when evaluating the
amplitudes detected.
So, is any of this information of use to
the analyst when he is trying to diagnose
bearing defect symptoms in a slow speed
environment well below the industry’s
advisory minimum speed limits? If the data
contains a repeatable signal source, with
multiple harmonics that exist at a frequency
spacing correlating with a non-synchronous
order that falls between 3 and 13 x RPM,
then there is a significant probability that a
bearing fault exists.
If we have used the 120 RPM speed
limit because the minimum frequency to
be detected is 2 Hz, and we ignore the
requirement to monitor the drive’s running
speed issues (imbalance, looseness,
misalignment, etc.) in favour of detecting
bearing failure, then as we are expecting
to see defects at 3 x RPM minimum, we
can be confident that we should be able to
detect faults at speeds down to 40 RPM. If
we further consider that we expect to see
multiple harmonics of the defect frequency,
then we can decrease the speed of the
rotating component even further.
When monitoring for these low
frequencies it should be remembered that the
amplitudes will be extremely low. However, it
should be remembered that if the alternative
is no monitoring at all because convention
says the data won’t be repeatable, what is
there to lose (apart from the failure)?
The following are real world examples
taken from my experiences in monitoring
low, variable and intermittent operation
production machinery in service. These
cases generally fall into speed and
operation categories that conventionally
dictate that the units are un-monitorable.
However, in each of these cases the fault
modes presented themselves readily to
either frequency spectra or time waveform
analysis, and have been performed as
part of routine maintenance programmes
at customer request in attempts to avoid
causing costly plant stoppages. In each of
these cases the concerns over speed and
sensor selections were explained to the
clients, and each application was trialled
fi rst to determine whether the programme
was likely to be successful.
ALUMINIUM REEL BEARINGS,
VARIABLE SPEED, 300 - 20 RPM
In this application, the bearings to be
monitored supported an aluminium rolling
mill re-coiler, which accepts thin rolled
sheet material at the output side of a rolling
reduction process. The speed of the strip
though the mill is an important parameter
of the quality of the rolled product, and
therefore to maintain the pre-set product
pass speed through the mill, the re-coiler
mandrel must reduce speed throughout
the coil build up, to compensate for the
increasing circumference of the built
coil. The load on the coil mandrel and
its bearings is a function of strip tension
and coil mass, the latter increasing as the
vol 23 no 3 maintenance & asset management | May/June 2008 ME | 31
coil diameter builds. Maximum speed for
the re-coiler mandrel is therefore at the
commencement of the coiling process, and
in this case was no more than 300 RPM
at the mill’s highest operating speeds,
dropping to 20 RPM as the coil fi nished.
The application of vibration monitoring
was required following catastrophic failure
of the mandrel support bearing races, which
disintegrated with no significant fore-
warning to production staff.
Vibration data was gathered using a
portable data collector, fi xed installation
standard accelerometers, and simultaneous
speed measurement to speed stamp
the gathered vibration data as it was
collected. To ensure maximum energy was
transmitted from any bearing fault, data
was gathered at the highest mandrel speeds
possible following the mill’s acceleration to
full rolling speed, and spectral collection set
to 1 average in an attempt to avoid smearing
of the frequency peaks as the re-coiler
decelerated to compensate for coil build up.
Data was gathered on a routine monthly
schedule during normal production,
and the subsequent analysis of the data
was undertaken manually off site. Each
vibration spectrum was manually checked
for the presence of abnormal mechanical
symptoms, and in particular the presence of
bearing defects.
Within two months of the bearing
replacement both roll support bearings
produced non -synchronous multiple
harmonics which correlated directly with
the bearing manufacturer’s predicted outer
race defect frequencies. As can be seen in
Figure 4, the amplitudes were extremely
low. The spectra showed, however, that
the fundamental defect frequency was
evident, indicating that the damage had
already breached the bearing surface. As
mentioned in the previous discussion,
defect frequencies should not be evident
at all in a new bearing (split race bearings
aside) and, as a consequence, lubrication
of the bearings was accelerated and
progression of the fault amplitudes
and frequency spread monitored for
deterioration. This didn’t occur for six
months, the increased lubrication managing
to hold further deterioration at bay, but
eventually increasing amplitudes at the
fundamental frequency commenced and the
bearings were replaced during a planned
maintenance stop.
 01 - Recoiler Mandrel Brg Fault
To ensure that the molten material does
RecM Brg -R2H Reel 2 NDE Brg Axial not adhere to the surfaces of the mould,
5
4
FAULT Trend Display the latter is agitated with a mechanical
RMS mm/Sec

Overall Value
3
2
ALERT
oscillating drive along the axis of the
1 > SK F 23 0 56C
C =B PF O product withdrawal route. If the oscillation
0
0 60 120 180 240 300 route or process is disrupted, or fails, then
0.04
Days: 01-Oct-03 To 08-Jun-04
the fragile cast skin of the product can
PK Acc in G-s

Analyze Spectrum
0.03 03-Dec-03 17:39:08 rupture, causing the molten product to
C C C C C C C C C C PK = .1593
0.02
LOAD = 100.0 escape, resulting not only in lost production
0.01 RPM = 231. (3.86 Hz)
0
and raw material for the strand concerned,
0 200 400 600 800 1000 but also the potential for expensive damage
Frequency in Hz
0.5 and downtime, depending on how severe
0.3 Analyze Waveform
the material breakout is.
Acc un G-s

0.1
-0.1
-0.3
-0.5
0 50
Figure 4
Figure 5
Examination of the bearing showed a
number of radial cracks in the race, mainly
clustered within the load zone, with some
cracks close to the race edge. The plant
were not concerned with further evaluating
the fault mode, but as can be seen in
the Figure 5 photograph very little wear
of the bearing surface had taken place.
However, the spacing of the cracks, and
their orientation to the rolling plane of the
bearing, suggested that they may have been
related to damage when the roll assembly
was in storage (false brinelling) or have
been produced during installation, and that
these minor damage sites were aggravated
in service by an ingress of the acid-based
rolling fluids in use on the mill. It was
further surmised that crack propagation of
the defects to the edge of the race may have
03-Dec-03 17:39:08
PK = .1643
PK(+/-) = .3946/.3800
CRESTF = 3.40
As a consequence, the mould oscillation
Freq: 46.96
can be monitored for evidence of changes or
100 150 200 250 300 350 400
Time in seconds
Ordr: 12.18
Spec: .00063
disruptions to what should be a smooth and
uniform mechanical motion. The motion
of the caster is frequently slow, with mould
oscillation varying between 60 and 200
RPM, and is monitored on many plants
with a dedicated displacement monitoring
device. This frequently forms part of the
process monitoring system, and monitors
the physical displacement of the mould
oscillation as a value, and also the mould’s
operating frequency. But these devices
are frequently overlooked as a means of
monitoring the drive health.
For many smaller casting plants, such
devices are not fitted, and in one case we
were asked to monitor the drive to assess
the reasons for mechanical damage being
caused to the oscillation drives. In this case
there were no displacement transducers
installed, and the plant considered the
cost of retro-fitting to be too high due to
the limited access available around the
machine. We also needed to verify whether
the technique of monitoring the oscillation
resulted in sudden break up of the outer of the drive and its linkage arms would
race with little physical warning. indeed be successful. So, having warned
the customer not to expect too much,
accelerometers were used with the intention
CONTINUOUS CASTING of evaluating the purity of the stroke as a
OSCILLATING DRIVE GEARBOX time waveform, and checking the purity,
– 90 RPM from one side of the table linkage to the
Continuous casting is a commonplace other, for discrepancies and differences in
method of producing metal forms or blanks amplitude.
for process purposes, and has replaced Initial traces were encouraging,
ingot casting routes for the majority of with newly overhauled tables producing
metal processing companies. The process sinusoidal motions at oscillating frequency,
is deceptively simple, with an open ended accompanied by a visible impact in the
mould fed continuously from the top with wave form as the drive backlash was excited
molten material, and a cast form withdrawn just after the bottom of the table stroke (see
from the bottom of the mould at a carefully Figure 6). This is produced as the friction
controlled, and sufficiently low, speed between the cast slab and the mould results
to ensure that the liquid material has a in a steady downward force on the drive
solidified skin before emerging. table and linkages – but at the reverse

32 | May/June 2008 ME | maintenance & asset management vol 23 no 3

Slow Speed Bearing Analysis – monitoring in the grey zone

of the stroke at the bottom of the travel, 82 - Con Caster Osc T1

gear backlash in the drive unit is excited. ContcstrT1-T1V Table Lift Brg

In addition, waveforms taken from both 0 .6 Route Waveform

11-Apr-07 16:45:14
sides of a table operating correctly should PK = .2698
produce symmetrical wave forms with 0 .3 LOAD = 100.0
RPM = 169. (2.82 Hz)
uniform displacement values.

Acceleration in G-s
PK(+) = .4385
The most common problems found PK(-) = .5934
CRESTF = 3.11
0
during the monitoring programme were the
development of impact forces immediately
prior to the table reaching the bottom of -0.3

the stroke (see Figure 7) or disrupted and

distorted wave forms, sometimes to one
-0.6
side of the table, producing asymmetric
results (see Figure 8). In most cases this was Normal Oscillation

produced by debris build up on the table’s -0.9 Time 7.371

Ampl: -.569
lower bump stops, or around the table’s 6 .2 6 .4 6 .6 6 .8 7 .0 7 .2 7 .4 7 .6 7 .8

edges and skirts, although failing linkage Time in Seconds

bearings could produce similar symptoms.

Figure 6
As a result of the work done to date,
debris cleaning operations around the
casting tables have been altered, and 82 - Con Caster Osc T1
ContcstrT1-T1V Table Lift Brg
targeted at specific areas found to be prone 0 .3 Route Waveform
to slag build up, and as a consequence table 24-May-07 09:52:53
PK = .0939
operation has improved in reliability. 0 .2
LOAD = 100.0
RPM = 61. (1.02 Hz)

PK(+) = .2207
PK(-) = .2517
Acceleration in G-s

CONTINUOUS CASTER 0 .1 CRESTF = 3.79

STRAIGHTENER SECTION ROLL

BEARINGS 0 .0

As an example of why monitoring

-0.1
should be experimented with, this fi nal
example just shouldn’t lend itself to
monitoring using conventional techniques -0.2

at all. A client asked us to develop a Table Bottoming on Debris Time 6.865

monitoring approach for their straightener -0.3 Ampl: -.252
4 .0 4 .5 5 .0 5 .5 6 .0 6 .5 7 .0 7 .5 8 .0
section roll bearings, which were failing Time in Seconds
after only six months of service. The roll
turns at just 4 RPM, creating concerns Figure 7
over whether the components would be
operating as conventional rolling elements.
The rolls involved transport cast billets 82 - Con Caster Osc T1
ContcstrT1-T1V Table Lift Brg
from a continuous casting machine to the 1.0 Route Waveform
cooling beds for distribution. The rolls are 09-Nov-07 10:25:68
0.8
water cooled but the bearings regularly PK = .3969
LOAD = 100.0
experience ambient temperatures of over 0.6 RPM = 193. (3.22 Hz)

100 degrees. Cost was an issue due to 0.4 PK(+) = .9177

PK(-) = .8758
Acceleration in G-s

the number of sensors required for full 0.2

CRESTF = 3.20

coverage, and their required locations meant

0.0
that accelerometers would be the only cost
effective measure, so a trial survey was -0.2

arranged during the plant’s maintenance -0.4

period. -0.6
With the rolls turning at only 4 RPM,
-0.8
analysis of the spectra was considered a Waveform distortion from excess Backlash in Lift Bearings
Time .926
tall order, but there was some hope that -1.0 Ampl: .08989
0 .3 0 .6 0 .9 1 .2 1 .5 1 .8 2 .1 2 .4
any significant defect within the bearing Time in Seconds
might produce either peak amplitude
changes in the time waveforms, or overall Figure 8

vol 23 no 3 maintenance & asset management | May/June 2008 ME | 33

 Slow Speed Bearing Analysis – monitoring in the grey zone

rises in carpet level for the time waveform.

Accelerometers were attached using
magnetic mounts, and as the data collector
being used is primarily set for spectral
analysis, the Fmax was set to 150 Hz at
3200 Lines, to produce a time waveform
of 20 seconds. This was not ideal, as a
longer time waveform would be required
to confi rm repetitious anomalies in the
time waveforms, however the monitoring
procedure would cause planning issues
for the maintenance stop, and time was
at a premium to survey the full set of rolls
targeted. However, if we did spot something
meriting further investigation we would
be able to spend the time gathering much
longer data sets as required.
I was therefore very surprised when I
examined the data from one of the rolls to
fi nd a series of harmonic frequencies at very
low frequency (see Figure 9). Once back on Figure 10
the computer the frequency matched the
predicted defect frequencies for the bearing
outer race, and arrangements were made CONCLUSIONS experimentation with available techniques,
to replace the bearing at a suitable stop to I am not advising that the particularly when the equipment to be
validate the data. recommendations of sensor and monitoring monitored is vital to manufacturing and
Inspection revealed that the outer race equipment manufacturers are ignored or is not monitored by other more viable
was spalled, with some cracking (see Figure discounted, as the wisdom passed down techniques and systems.
10), indicating that the survey had been through training has solid grounding in However, once the initial investment in
able to detect an early Stage 3 bearing defect the physics of vibration monitoring and monitoring equipment has been made, there
at frequencies well below the accepted equipment limitations. I am also not is little to be lost in trying to push the limits
limits for the sensor and the data collector. suggesting that conventional early warning as far as possible, particularly when the
At present this survey is being conducted bearing techniques should not be used alternative might be no monitoring at all. In
on a routine basis during maintenance in conjunction with spectral and time the examples discussed above, I have simply
stoppages. waveform analysis, but I am advocating looked for frequencies or anomalies which
shouldn’t be present in a healthy machine,
82 - Strand Straightener Rol
and taken little consideration of the overall
Str Roll -R1H Roll Brg Horizontal amplitudes as a measure of health.
0 .020
As ever, failure investigation to evaluate
RMS Velocity in mm/Sec

Analyze Spectrum
EE E E EE E EE E 06-Sep-07 11.56:50
0 .015 RMS = .0800 the success of the programme must take
LOAD = 100.0

0 .010
RPM = 4. (.06 Hz) place. If the component is replaced too early
>SKF 23122CC in its fault progression it is still a fault saved,
E=BPFO
0 .005
and the information gleaned lends support
0 to the concept that the fault is detectable.
0 5 10 15 20 25 30 35 40 This enables the analyst to assess the
Frequency in Hz
0 .02 Analyze Waveform
symptoms and allow for further service
life at a subsequent detection. If the fault
Acceleration in G-s

06-Sep-07 11:56:50
0 .01 PK = .0032
PK(+/-) = .0109/.0177 is missed, and cannot be seen in the data
CRESTF=7.81
0 even with hindsight, then if the alternative
-0.0 1
was not to apply monitoring, nothing was
lost in the effort. Nothing ventured, nothing
-0.0 2
Freq: .604 gained!
0 10 20 30 40 50 Ordr: 9.787
Spec: .00639
Time in Seconds
The author may be contacted at:
info@amsolve.co.uk
Figure 9

34 | May/June 2008 ME | maintenance & asset management vol 23 no 3