VIBRATION MONITORING

OF GEARBOXES

Written By
James C. Robinson,
Technical Consultant, IMI division of PCB Piezotronics

Curated By
Meredith Christman,
Product Marketing Manager, IMI division of PCB Piezotronics

pcb.com | 1 800 959 4464

VIBRATION MONITORING OF GEARBOXES

Article by:
James C Robinson, Technical Consultant, IMI division of PCB Piezotronics

Curated by:
Meredith Christman, Product Marketing Manager, IMI division of PCB Piezotronics

INTRODUCTION

Gearboxes are a common component in many industries, including chemicals, sugar, steel, mining, plastics, oil &
gas, power and petrochemical. Because their operation is essential to many manufacturing processes, planned
downtime is inconvenient and unplanned downtime can be catastrophic.

The objective of this paper is to demonstrate a wide variety of fault types and the importance of employing
proper sensors and analysis tools. Particular emphasis is placed on vibration monitoring employing
accelerometers as sensors on industrial gearboxes.

SECTION 1: SENSOR SELECTION AND MOUNTING

Potential Faults of Gearboxes: Prior to selecting sensors and addressing their mounting on a gearbox, it will first
be beneficial to remind ourselves of the characteristics of the vibration signature that can be expected from
common gearbox faults. For example, if there is a crack in a gear, it will likely introduce a slight speed change
when the defective tooth enters the load zone. This leads to impacting every time that tooth enters into its load
carrying responsibility (typically once per revolution of that gear). Now consider a situation where there is a lack
of sufficient lubrication for the gear teeth going into and out of the load zone. This leads to friction between the
teeth with the maximum activity typically occurring twice per tooth mesh (once on the addendum and once on
the dedendum). Of course, both friction and impacting can also occur within the gearboxes. In gearboxes,
impacting will generally be periodic and friction generally non-periodic (random).

If a fault is present that is generating stress wave activity on a set of meshing gears, that energy will be
transmitted to the outer housing via the shafts where the gear set is attached through the bearings (providing
they are rolling element bearings). An accelerometer fastened to the outer surface in the proximity of that
bearing would capture the stress wave activity providing the accelerometer has sufficient bandwidth and
sensitivity. If the bearing is a sleeve bearing, significant attenuation will occur to the stress waves in coupling
across the gap from inner race to outer race and thus may not be sufficient for capture by the sensor. A
proximity probe lacks sufficient sensitivity for the relatively high frequency stress wave activity.

In addition to the relatively high frequencies present in the stress wave packets generated by friction and
impacting, the lower frequencies generated by faults such as misalignment, looseness and balancing issues must
also be captured and analyzed. These stress wave packets contain frequencies in one of the following two
ranges:
• About 0.3 times running speed to about 3.25 times the gear meshing frequency.
• About 0.3 times running speed to about 50 times running speed.

The analysis of the low-frequency band is carried out by capturing a time waveform and proceeding to
transform that time waveform into spectra data on which most of the diagnostics are carried out. For the high-
frequency band, the most common procedure is to run the signal from the sensor through a high-pass filter
followed by full-wave rectification. The rectified signal is then demodulated to extract any periodic or random
1
activity that is occurring. If periodic or random activity is occurring, the analyst needs to know the periodic rate
as well as the amplitude of the activity.
Classic demodulation techniques do not
maintain the true g-level. The cases
presented in this paper use the PeakVue™
methodology developed by Emerson Process
Management, which does maintain the true
peak g-level.

When the analyst does find a fault in a piece

of equipment and reports it to operations,
the typical response from operations is to
ask about the severity of the fault and how
long can the equipment continue to operate.
These questions sometimes are difficult to
answer. For the low-frequency band data,
there are several charts available in the
industry that can be used to make intelligent
judgment calls on both fault severity and
continued equipment operation. For the
high-frequency band, assuming the true peak acceleration level and
speed (RPM) are available, the chart in Figure 1 has been created to
provide similar intelligent answers on both fault severity and
continued equipment operation. The level given in the chart is the
recommended alert level; the recommended fault level is twice the
alert level.

ICP® Accelerometer Use: The most common sensor type employed in

vibration analysis on gearboxes are ICP® accelerometers with a
sensitivity of 100 mV/g, a resonant frequency in the 25 kHz range and
a noise floor of approximately 100 μg/√Hz at 1 Hz (or less). IMI
Model 603C01 (top exit with ¼-28 female mounting thread) would be
an example of an ideal model. The specification sheet for the
accelerometer typically specifies the sensitivity is nominally flat to

within 3dB from a fraction of 1 Hz to 10 kHz. See

Figure 2 for an example of the characteristic
compliance of Model 603C01.
The implicit assumption is that the sensor is
attached to a clean flat surface with a stud at a
specified torque. Because stud mounting is both
expensive and time consuming, its requirement
encourages sparse data acquisition. The analyst
will often turn to a much simpler means of
attaching the sensor to the surface, such as using
a two-rail magnet placed on a curved surface with
the sensor attached to the magnet. This approach
will often lead to not capturing the higher
frequencies associated with impacting or friction.

2
To explore the impact that sensor mounting has on the sensor frequency response, frequency response data
was captured (presented in Figure 3) for a sensor that was:
• Mounted with a stud with grease on a flat dry surface.
• Mounted with a stud without grease on a flat dry surface.
• Mounted with a flat magnet on a flat clean surface.
• Mounted with a dual-rail magnet mounted on smooth curved surface.
• Mounted with a dual-rail magnet mounted on rough curved surface.
• Mounted with a dual-rail magnet mounted on painted curved surface.

For faults that manifest themselves in the frequency range of less than 2 kHz such as alignment, unbalance and
looseness, the results would be independent of how the sensor is mounted. For faults identified with higher
frequencies (impacting and friction), the results would be highly dependent to how the sensor is mounted
ranging from no response to distorted response.

Of course, the best way to mount would be stud-mount with a specified torque. This could get expensive as it
requires a dedicated accelerometer at every measurement point. An acceptable alternative mounting would be
to use a flat magnet placed on a flat smooth surface such as a mounting pad. The flat magnet approach to all
measurements points combined with stud mounting the sensor in radial direction on the inboard and outboard
ends would be a recommended
method for sensor mounting.

To illustrate the type of effect that

sensor mounting can have on friction
activity, data is presented in Figure 4
from a case where it was known that
bearing lubrication was needed. In the
lefthand time trace, the sensor was
mounted using a flat magnet attached
to a flat smooth surface. A second set
of data on the righthand time trace
was acquired from a sensor attached
to a curved surface via a dual-rail
magnet. Both time traces were taken at the
same time using a two-channel data
collector. The bearing was lubricated
(greased) at the time the sudden level of
noise decreased.

SECTION 2: COOLING TOWER GEARBOX

A metals plant lost the use of one of its

cooling towers due to a two-speed gearbox
becoming unusable. The gearbox
replacement had a long lead-time but
sufficient parts were available to put
together a temporary gearbox replacement.
When the cooling tower was placed back in
service with the temporary gearbox at low
speed, vibration velocity spectral data was

3
acquired and is presented in Figure 5. The peak-to-peak g-level is less than 10 g’s, and the spectral data is a “ski
slope”. This is typical for an ICP® accelerometer that is being overloaded. The sensor was a 100 mV/g
accelerometer with a 9-11 VDC bias voltage.

The accelerometer was replaced with a 10

mV/g accelerometer with a similar bias
voltage. Using the 10 mV/g sensor, a 40 kHz
acceleration spectra data set with
waveform was captured and is presented as
Figure 6. The peak-to-peak g-level was close
to 200 g’s, explaining why the 100 mV/g
sensor’s output was unstable. The primary
cause of the high g-levels was friction
occurring within the bearings as well as
between the gear teeth as a result of an
inoperable lubrication system driven by a
gear oil pump.

The operators chose to continue operating

the cooling tower rather than take the time
to correct the lubrication problem. They did
ask the analyst if it would be best to run the gearbox at high speed or at low speed. It was decided to run high-
frequency analysis at both speeds to see if the best speed could be determined from the data. The high-
frequency band chosen for the analysis was 2-40 kHz. The results for the gearbox running at high speed are
presented in Figure 7. There was significant periodic activity at one and two times gear mesh frequency. The
peak g-level in the waveform is 150 g’s, which is very high. The results for the gearbox running at low speed are
presented in Figure 8. The significant difference between the high-speed and low-speed activity is the lack of
any gear mesh frequency or twice gear mesh frequency activity in the high-speed data relative to the low-speed
data. The peak g-level in the high-speed data is 140 g’s, which is slightly lower than the peak g-level of 150’s in
low-speed data. The presence of the gear mesh frequency and twice gear mesh frequency in the low-speed data
suggests that debris from the friction between gear teeth was being thrown out into bearings. The conclusion
was that there was sufficient oil in the bottom of the gearbox to permit the gear to sling oil out to lubricate the
gear teeth when running at high speed. Therefore, it was decided that less damage would occur with the
gearbox running at high speed. The gearbox was run at high speed until replaced with the new gearbox.

4
SECTION 3: MULTI DRILL HEAD GEARBOX

Vibration measurements were taken on

several multi drill head gearboxes at an
automobile transmission plant. The primary
objective was to demonstrate that vibration
analysis of the high-frequency band could
reliably detect early-stage bearing faults in
drill heads. In addition to bearing faults,
several other faults were found. One of
those other faults was a torsional vibration
problem on a particular gearbox with ten
drill heads. The gear arrangements were
accomplished with the two levels depicted in
Figure 9. Assuming the two-pole motor was
running with no slip (3,600 RPM or 60 Hz),
the gear mesh frequency for the level 1 gear
set is 2,280 Hz. Three times gear mesh
frequency is 6,840 Hz, thus the use of a low-
frequency bandwidth of 20-8,000 Hz was
most appropriate. There are possibly three
additional gear mesh frequencies associated
with the gear configurations in the level 1
configuration presented in Figure 9. The gear
mesh frequency for the cluster of gears
driven by I-1 is 2,146 Hz. The other two gear set
clusters driven by I-3 and I-4 are 2,280 Hz.

The low-frequency band spectra and waveform

data are presented in Figure 10. The peak-to-peak
g-level in the waveform data is about 14 g’s. There
is definite periodic activity around 1,000 Hz. There
are probably two individual gear mesh frequencies
in the expected frequency range of 2,100 to 2,300
Hz. Additionally, there is a broad band of spectra
activity in the 2,500 to 2,800 Hz frequency range.
The high-frequency (5 to 40 kHz) rectified
waveform and spectra data taken at the same time
as the low-frequency band data are presented in
Figure 11. The spectra bandwidth of the high-
frequency data in Figure 11 was specified at 1,000
Hz, which was an oversight. It should be slightly
greater than twice gear mesh frequency (5 kHz). The
peak g-level in the rectified waveform is about 8 g’s. The
bothersome feature of concern in the high-frequency
spectra data in Figure 11 is the activity at slightly greater
than 0.5 times running speed. It is an indication of a
possible torsional resonance. Based on this observation,
it was decided to inspect the motor shaft/key. A picture
of the motor shaft/key is presented in Figure 12. There
5
has been considerable wear in the motor shaft keyway. The recommended solution was to decrease the
tolerances in the keyway fit to the shaft. This corrected the shaft/key wear problem.

Spectra data taken after the motor shaft replacement is presented in Figures 13 and 14. The resolution is
sufficient to clearly show the gear mesh frequency at 2,275 Hz as well as a probable torsional resonance at a
frequency slightly greater than gear mesh frequency. The periodic activity that was seen in the high-frequency
rectified waveform data before motor shaft change is gone. Additionally, the periodic activity that was seen at
about 1,000 Hz in low-frequency spectra data before motor shaft change is also now absent. The most obvious
change between the two sets of data is the significant increase in the vibration waveform amplitude. The peak-
to-peak waveform in the low-frequency data went from 14 g’s peak-to-peak to 40 g’s peak-to-peak. A similar
change occurred in the high-frequency rectified waveform. This is probably because the torsional stiffness
increased with the tightening of the system at the motor shaft/key thereby increasing the torsional resonance
frequency. The increase in vibration is more desirable than the shaft/key interaction.

SECTION 4: CRUSHER GEARBOX

This is a large gearbox measuring roughly 8 x 10 x 20 ft

used for crushing rocks at a mining facility. A plan view
of the gearbox is presented in Figure 15. There are
twelve vibration-monitoring points identified in Figure
15 as 1 through 12 which are used for scheduled
acquisition of vibration data.
Routine monthly vibration data was acquired from all
the measurements points on the gearbox. The high-
frequency spectra data and rectified waveform from
measurement point #2 are presented in Figure 16. The
high-frequency band is from 1 kHz to 40 kHz. The 1 kHz
is well above 2.25 times gear mesh frequency. (Gear
mesh frequency is 330 Hz, which equals running speed
of motor [15 Hz] times the number of teeth [22]). The
bearing has an inner race fault in the early stages of
failure based on relatively low peak g-level of 3.6 g’s.
The measurement was repeated approximately one
month later and results are presented in Figure 17. The big difference in Figures 16 and 17 is the peak g-level has
increased from 3.6 g’s to 51 g’s, which is significant. This bearing, based on high g-level of 51 g’s, requires
frequent monitoring with a replacement plan set in motion.

6
To assure the faulty bearing was bearing #3 and not bearing #1 or bearing #2, data was acquired from
measurement point #1 at same time as the data in Figure 17 was acquired. The peak g-level in the high-
frequency data from measurement point # 1 was 25 g’s versus the 54 g’s from measurement point #2. The
defective bearing was concluded to be bearing #3. Frequent monitoring with the sensor set on measurement
point #2 was then carried out. The peak g-level trend is presented in Figure 18. From the time the 50 g peak
level was detected, it took about 25 days to get the bearing replaced. A picture of the defective bearing is
presented in Figure 19. The peak g-level recorded was about 80 g’s.

SECTION 5: PRECISION TENSION BRIDLE GEARBOX

A plan view of a precision tension bridle gearbox is presented in Figure 20.

The gearbox has a single shaft input with a dual shaft output that ultimately
drives the work rolls. The low-frequency and high-frequency band spectra and
waveform data were taken with sensor placed over bearing at input to
gearbox. The low-frequency band data is presented in Figure 21. The input
gear mesh frequency (approximately 351 Hz with second harmonic) is being
modulated. The lower gear mesh frequency (approximately 202 Hz) for the
90-tooth gear set is sharp (no indication of modulation) and the second
harmonic is not discernible. The rectified waveform and spectra data for the
high-frequency band of data from 0.5 to 40 kHz are presented in Figure 22.

7
The spectra data indicates a significant event is occurring once per output shaft revolution. The waveform
clearly demonstrates there are two distinct impulses per turn that could mean there are two cracks on one gear.
When inspected, two cracked teeth on one gear were found.

SECTION 6: POST REBUILT GEARBOX WITH CRACKED TOOTH

A gearbox was removed from a mining machine and sent

back to the shop for rebuild. A plan view of the gearbox is
presented in Figure 23. This gearbox has four gear mesh
frequencies. After the rebuild, the machine was subjected to
vibration analysis for high and low-frequency bands prior to
sending it back to the field. The vibration analyses are carried
out on the twelve measurement points (as needed) identified
on the gearbox as points G1 through G12. There is a
measurement point placed over each bearing. In this case,
focus is directed to the low-frequency band up
to 1.5 kHz and the high-frequency band from 1
kHz to 40 kHz data from measurement point
G5.

8
The waveform and spectra data for the low-frequency band are presented in Figure 24. The peak-to-peak g-level
in the waveform data is about 1.1g’s, which is not considered to be indicative of a problem. The high-frequency
band rectified waveform and spectra data are presented in Figure 25. There is a spike (about 14 g’s) occurring
once per revolution of the shaft immediately below measurement points G5 and G6. This is the expected
signature from a cracked tooth in the gear immediately below measurement point G5. Upon inspection of this
gear, it was found to have a crack in the root.

SECTION 7: POST REBUILT GEARBOX WITH ECCENTRIC GEAR

A gearbox was removed from service and sent

to the shop for rebuild. Upon completion of
the rebuild, the gearbox was subjected to low
and high-frequency vibration analysis. The low-
frequency measurement showed no sign of a
problem. The high-frequency band (from 0.5 to
40 kHz) rectified waveform and spectra data
are shown in Figure 26. The high-frequency (0.5
to 40 kHz) rectified waveform is showing a
definite repeating pattern of increased activity
over approximately 65% of a single revolution

and low amplitude activity over the

remaining 35% of the revolution. This is the
pattern expected for an eccentric gear. In
the spectra data, there is significant activity
at gear mesh frequency, which is indicative
of a stable rotational speed of the gear.
Following replacement of the eccentric gear,
the high-frequency rectified waveform and
spectra were acquired and are presented in
Figure 27. Someone was not convinced
the gear was eccentric and placed the
defective gear in another gearbox. When
it was subjected to testing, the results
presented in Figure 28 were obtained
which was basically the same as that
presented in Figure 26.

SECTION 8: FATIGUING IN BEARINGS

AND GEARS

Many failures in bearings and gears are

initiated by residual stress building up in
the metal (e.g., bearings or gears) under
usage. When the sum of residual stress and current usage stress exceeds certain levels, the residual stress will
be relieved by cracks (fatiguing) starting beneath the surface and proceeding to the outer surface as the cracking
progresses under use. When stress relief cracks are initiated, they are accompanied with the emission of stress
wave packets that travel at the speed of sound in the material to the outer edge of the component and away
from the initiation site. The frequency within the packets is similar to packets emitted when friction occurs (ie.

9
higher frequencies than when impacting
occurs). To illustrate this type of failure, a case
from a 1.5 MW wind turbine generator gearbox
has been chosen for illustration. The gearbox is
made up of a three-wheel planetary gear
section driving a two-stage spur gear section.
Both sections were housed in a compact
housing. An online continuous vibration
monitoring system was monitoring the
gearbox.

The rectified high-frequency band (2-40 kHz)

waveform and spectra data are presented in
Figure 29 from a sensor mounted over the
outboard spur section of the gearbox. The
spectra data is dominated by activity at 5.253
orders with many harmonics. This is the outer
race fault frequency for the bearing used on the
outboard end of the intermediate shaft in the
spur gear section. The observed peak g-level of
1.15 g’s for this speed shaft (538 RPM) is very
low. The recommended alert level is 4-5 g’s. This
fault could very well be from the early stage
fatiguing with cracking occurring under the
outer race surface. The fault may not be visible
if examined at this time by sight or by feel.
Typically, wait until peak g-level reaches 6-8 g’s
(for this speed machine) before a plan is put in
place for changing out the bearing unless there
are special circumstances such as it is most economical to implement the change at this time.

The data in Figure 30 was chosen to illustrate how the fault signature can vary over reasonably short times. Each
rectified time waveform is for a period encompassing 18 revolutions (2 seconds) of the intermediate shaft
turning at 538 rpm. The last trace of data in Figure 30 is the same set of rectified waveform data presented in
Figure 29. The three remaining data traces were acquired one day previous.

There is very little fault signature showing up in the first rectified waveform data. In the middle two traces of
rectified waveform data, a fault signature is present but with considerable variability. If the signature was a
result of impacting from a defect in the inner surface of the outer race, the rectified time waveform would not
be expected to have the variability seen in Figure 29.

SECTION 9: CONCLUSIONS

Several faults common to gearboxes were presented in this paper. The objectives were to:
• Demonstrate a wide variation in fault types.
• Demonstrate the importance of employing proper sensors and analysis tools.

Many faults generate a short burst of stress wave activity (from impacting and friction) that requires sensors
responsive up to 15-25 kHz, such as IMI Model 603C01. The sensor frequency response is strongly dependent on
how the sensor is attached to the surface of the machine. Accelerometer mounting by stud, by flat magnet on a
10
flat smooth surface and by dual-rail magnet on a curved surface (smooth, rough or painted) were all considered.
The conclusion was that the stud-mounting technique was the preferred choice. Mounting the accelerometer
with a flat magnet attached to a smooth, flat surface provided sufficient bandwidth for impacting and friction
detection. The use of a two-rail magnet on a curved surface would miss several situations where impacting and
friction was occurring.

The analysis tools used in this study consisted of the normal spectra data in velocity units and waveform in
acceleration units. To cover the high frequency burst of stress waves from impacting, friction and fatiguing, the
waveform used was the band-limited rectified signal. The band-limited rectified signal was 0.5-40 kHz, 1.0-40
kHz or 2.0-40 kHz. The rectified waveform was also transformed to spectra data in acceleration units. In addition
to waveform and spectra data, the autocorrelation coefficient data was computed from the waveform data
graphically displaced. It was a very useful diagnostic tool in both fault identification and severity assessment.

11
 IMI Sensors, a division of PCB Piezotronics, Inc. manufactures industrial vibration monitoring
instrumentation, such as accelerometers, vibration transmitters and switches that feature rugged
stainless steel housings and survive in harsh environments like paper and steel mills, mines, gas
turbines, water treatment facilities and power plants. Integrating with portable analyzers and
PLC’s, IMI instrumentation helps maintenance departments reduce downtime and protect critical
3425 Walden Avenue, Depew, NY 14043-2495 USA machinery. Visit IMI Sensors at www.pcb.com. PCB Piezotronics, Inc. is a wholly owned subsidiary
Toll-Free in the USA: 800 959 4464 of MTS Systems Corporation. Additional information on MTS can be found at www.mts.com.
Phone: 1 716 684 0001 | Email: [email protected]
© 2019 PCB Piezotronics, Inc. In the interest of constant product improvement, specifications are subject to change without notice. PCB ®, ICP ®, Swiveler ®, Modally Tuned ®, and IMI ® with associated logo are registered trademarks of PCB
Piezotronics, Inc. in the United States. ICP ® is a registered trademark of PCB Piezotronics Europe GmbH in Germany and other countries. UHT-12 TM is a trademark of PCB Piezotronics, Inc. SensorLine SM is a service mark of PCB Piezotronics. Inc.
SWIFT® is a registered trademark of MTS Systems Corporation in the United States.
WPL_82_0219

MTS Sensors, a division of MTS Systems Corporation (NASDAQ: MTSC), vastly expanded its range of products and solutions after MTS acquired
2 PCB Piezotronics, Inc. in July, 2016. PCB Piezotronics, Inc. is a wholly owned subsidiary of MTS Systems Corp.; IMI Sensors and Larson Davis are
divisions of PCB Piezotronics, Inc.; Accumetrics, Inc. and The Modal Shop, Inc. are subsidiaries of PCB Piezotronics, Inc.