UNCLASSIFIED

Defense Technical Information Center

Compilation Part Notice
ADP013505
TITLE: Detection and Severity Assessment of Faults in Gear Boxes from
Stress Wave Capture and Analysis
DISTRIBUTION: Approved for public release, distribution unlimited

This paper is part of the following report:

TITLE: New Frontiers in Integrated Diagnostics and Prognostics.
Proceedings of the 55th Meeting of the Society for Machinery Failure
Prevention Technology. Virginia Beach, Virginia, April 2 - 5, 2001
To order the complete compilation report, use: ADA412395
The component part is provided here to allow users access to individually authored sections
)f proceedings, annals, symposia, etc. However, the component should be considered within
[he context of the overall compilation report and not as a stand-alone technical report.

The following component part numbers comprise the compilation report:

ADP013477 thru ADP013516

UNCLASSIFIED
 DETECTION AND SEVERITY ASSESSMENT OF FAULTS IN GEAR BOXES
FROM STRESS WAVE CAPTURE AND ANALYSIS

James C. Robinson
Computational Systems, Inc.

Abstract: Many faults in gearboxes are accompanied by the emission of stress waves
that disperse away from the initiation site at the speed of sound in metal. The wave
propagation introduce a propagating ripple on the surface which will introduce a
response, output, in a sensor sensing absolute motion such as an accelerometer. For an
accelerometer at a fixed location, the wave propagation will be a reasonable short-term
transient event lasting on the order of fractional to several milliseconds. The duration of
the event will be dependent on (1) type of event e.g., stress waves from impacting will
last longer than stress waves accompanying the release of residual stress buildup through
fatigue cracking, (2) relative location of the sensor (accelerometer) to the initiation site,
and (3) severity of the fault responsible for the stress wave emission.

For a healthy smooth running machine (gearbox), there generally will be no stress waves
present. Therefore their presence is indicative of a defect which generates stress waves.
Some common defects which generate stress waves are pitting in the races causing the
rollers to impact, fatigue cracking in bearing raceways or gear teeth (generally at root),
scuffing or scoring on gear teeth, cracked gear teeth, and others. The challenge becomes
one of detecting and quantifying relative to energy and repetition rate (or lack thereof) the
stress wave activity. This leads to the identification of certain faults and, with experience,
their severity.
The methodology employed by CSI per the capture and analysis of stress waves are to
collect a block of data consisting of peak values (in g's) which occur within discrete
sequential equal time intervals determined by the resolution sufficient to identify faults.
The number of time intervals over which peak values are collected are consistent with
that needed to invoke spectral analysis for the desired resolution and spectral band width.
The magnitude of the stress wave packets is identified in the discrete time data block
containing peak values. The presence of periodicity is identified in the spectral
(frequency) domain. An alternative to spectral analysis for the identification of
periodicity is auto-correlation analysis.
To illustrate this peak value (PeakVue TM ) analysis for fault detection and severity
assessment in gearboxes, several case studies are presented. The specific faults
demonstrated are bearing faults, cracked gear teeth, unstable driver speed, and torsional
vibration. It will be demonstrated that the PeakVue methodology is a very beneficial tool
for monitoring gearboxes.

Introduction: Many mechanical faults within industrial rotating machinery manifest

themselves through modal excitation (vibration) and stress wave initiation. Modal
excitation can be detected using sensors which detect absolute or relative motion. A
common sensor employed for detection of absolute motion is the accelerometer. The

329
The method employed by CSI for stress wave analysis avoids the use of a low pass filter
completely. This is accomplished by separating, as much as possibly, the stress wave
activity (the short term transient activity) from the continuous activity by routing the
signal from the sensor through a high pass filter set consistent with possible fault
frequencies within the machine. The resultant time signal is converted into a digital signal
at constant time increments for further analysis. The digital value recorded over each time
increment is not the signal value at a specific time, but instead is the absolute maximum
(peak) value observed over each discrete time interval. The resultant digital
representations are peak values, which occurred over each time increment.
The analysis of the peak value (PeakVueTM) waveform is basically (1) the identification
of any periodic activity occurring at rates consistent with possible fault frequencies and
(2) severity of assessment based on the level (peak value) of the stress wave activity. The
presence of periodic activity is identified through spectral analysis of the digital block of
data consisting of the sequential peak values. Severity level is extracted from observed
peak values compared with similar faults and/or trending of the peak values.
3. Case Studies
3.1 Introduction: The case studies chosen for presentation will demonstrate an outer
race and an inner race defect in separate pinion stand gear boxes. Sufficient data was
available to demonstrate the importance of trending for the inner race defect case. The
third case demonstrates severe cracked teeth in a Precision Tension Bridle gearbox. The
fourth case is from an extruder gearbox being driven by a DC motor with speed variation.
The fifth case was selected to demonstrate a torsional resonance problem present in a
large crusher gearbox.
3.2 Outer Race Defect in Pinion Stand Gearbox: This pinion stand gearbox was
included in the scheduled monthly condition monitoring program employing vibration
analysis. The traditional vibration monitoring showed no indication of a bearing fault. In
July 1997, the PeakVue methodology was introduced into the monitoring program.
It was obvious from the PeakVue data there was an outer race defect on the inlet shaft.
The peak g readings were 18 g's (the normal vibration readings were showing 1.5 g's
with no indication of a problem). The peak g readings in PeakVue continued to trend up
(got to 38 g's in mid Sept. 1997) and then started a downward trend (14 g's in early
October). The bearing was then replaced the peak g-levels on the replacement bearing
was less than I -g.
The normal vibration spectra and acceleration time waveform for data acquired on
September 15 1997 are presented in Figure 1. There is some indication of a possible outer
race problem but not conclusive.
The data from the PeakVue methodology acquired on the same date are presented in a
Figure 2. The peak absolute g-levels are up to 38 g's with a recurring rate consistent with
the outer race defect frequency. The bearing was replaced in early October 1997. A
picture of the defective bearing is presented in Figure 3. The defects in the outer race are
apparent.

330
analysis of the modal motion relative to the machine health is referred to as vibration
analysis. The methodology employed in vibration analysis consists of:
1. Capture (digitally) of a time waveform from a sensor for a specified time period.
The signal is first passed through a high order low pass filter prior to
digitalization. The purpose of the low pass filter is to remove all frequency
content which exceeds the Ngquist frequency (one half the sampling rate).
2. Transform the modified discrete time waveform into the frequency domain
employing FFT methodologies.
3. Look for excessive activity compared to other similar machinery or previous
history at discrete know fault frequencies.
The implicit assumption in vibration analysis is the signal being analyzed is stationary
equilibrium. The spectral values are average values, which are appropriate for stationary
(continuous) conditions.
Stress waves in metal accompany actions such as impacting, fatigue cracking, scuffing
(scoring) abrasive wear, etc. Stress wave emissions are short term, lasting several
microseconds to a few milliseconds, transient events which propagate away from the
initiation site as bending (s) and longitudinal (p) waves at the speed of sound in metal.
The s waves introduce a ripple on the surface which will excite an absolute motion sensor
such as an accelerometer. The detection and classification of these stress wave packets
provide an important diagnostic tool for (a) detecting certain classes of problems and
(b) severity assessment.
In the next section, a brief discussion of the methodology employed by CSI for stress
wave capture and analysis will be presented. The next section will present several case
studies showing the detection and severity assessment for faults commonly experienced
in gearboxes. The last section will be a summary and conclusion section.
2. Capture and Analysis Methodology for Stress Wave Activity: Stress waves are
generated when impacting, scuffing (scoring), fatigue cracking, abrasive wear, etc. are
present The duration of an individual event will range form fractional to several
milliseconds. The rate at which individual events occur within rotating machinery
generally are periodic consistent with the fault, e.g., a pitted area in the outer race will
cause impacting at the outer race fault frequency.
The sensor generally employed for the detection of stress waves is the accelerometer.
Since the signal is a short-term event relative to the repetition rate, the methodology
employed for the detection of the events preferable should avoid any averaging. This is
because there can be a large variation in repetition rate and hence the duty cycle will
introduce large variations in average values independent of the severity level of the fault.
Averaging negates the ability to perform severity level analysis based on trending and
relative comparisons.
When executing vibration spectral analysis, the general procedure is to route the sensor
analog signal through signal conditioning, which includes a low pass filter (anti-aliasing)
immediately prior to conversion into a digital signal with discrete values at a constant
sampling rate. The low pass filter is an averaging process; hence any short term events
are averaged over the averaging time associated with the anti-analyzing filter.

331
 "0.02 0 SPECRUM
TT

I :: 700 - -
e.Z.,
L. -1

,15t.3 7 10:4S
1
. "IS.I7

CR0077-
424

o 100 457 400 400 255 Moo TOO MSo ERr 240

Figure 1. Vibration velocity spectra and acceleration time waveform on pinion stand
gearbox on September 15, 1997.
GReO41-t
4TPO-AdTMSOPIlNION•
0.2 - 173p
WEST
4T57PIN005.53V SHAFT EYE050VYYT
MILL

1,t2 RMS-2,8

SR.. .

II I.3 i~i 40F-....L

lilA LT. HO

44 0A150044
YSPIS
0
CRET0--100

o o
S I I c .0.12
Pcoo •

RIiTI M,- 7S

Figure 2. PeakVue spectra and time waveform on Pinion stand gearbox (same as Fig. 1.)

Figure 3. Defective bearing taken from the pinion gearbox of Figs. 1 and 2.

332
3.3 Inner Race Defect in Finish Mill Pinion Stand Gearbox: This pinion stand
gearbox is separate from the example presented above. A separate data point was set in
the database and data (PeakVue and normal) acquired on a scheduled basis beginning on
Mar. 16, 1998. One of the trend parameters captured for trending was the peak g-levels in
the PeakVue time waveform. Experience has shown this to be a key parameter for fault
detecting and severity assessment.

The PeakVue peak g-level trend parameter for the lower output shaft and PeakVue
spectra for last collection date of May 28, 2000 are presented in Figure 4. The alert and
fault levels are set at recommended levels for this speed machine and type fault. From the
spectra, the fault is an inner race fault which is side banded (amplitude modulated) at
running speed which is indicative of fault going in and out of load zone at running speed.
From Figure 4, it is obvious that the fault exceeded the "fault" level about 7 months prior
to replacement, in July 2000.

The trend value for bearing fault over the same time interval for the normal vibration
monitoring are presented in Figure 5. Here there are no indications of a bearing fault.

Based on the trend values in PeakVue, a work order was release in June 2000 to replace
the bearing. The bearing was replaced in July 2000. A picture of the defective bearing is
presented in Figure 6. The failure was clearly advanced and could have induced
catastrophic failure easily by e.g., metal "chunks" interfacing with the gear teeth
meshing.

3.4 Cracked teeth in a Precision Tension Bridle Gearbox: This gearbox was a single
speed reduction gearbox with a dual shaft output. The slow speed shaft from the
reduction gear set (40 teeth pinion gear driving a 158 teeth bull gear) was driving a
second output shaft through a dual 90 tooth gear set. The input shaft was turning at
525 rpm and the output shafts turning at 215 RPM at the time the data presented below
was acquired.
20
STAND
PINION
. MBI[O0
FpgHOT
HM
. . ............ ........... ............ ............ ............ ........... - ....... . .......... -

°* A... . ..T4.0.....
.t
A ............ ........... ............ i.............) ........... --........... .......
------- . . ......... o t t-.r.e
S ............ •.. .... .............
............ "............
• .......... -- . .....• ..........

0.12
I, 1itT-T0
r~e
0940.0.01
ioasa i~

TO Ol-M:3O

P- HP500HT)
OWPALL--1 •
RMS
- .4024
.OAD
0F24
R 1.00

0.
o

Figure 4. Maximum peak g-level (from PeakVue) trend from March 16, 1998 to May 25,
2000 and PeakVue spectra from May 25, 2000.

333
 FAULT . 00000R0

.............
.. ........ .. ........... .....
........
...
................... ...
... ... OOM-S.O
011 0

Figure 5. Normal vibration bearing fault trend and spectra for latest measurement over
same time period as in Fig. 4.

Figure 6. Defective bearing taken from the pinion stand gearbox of Figs. 4 and 5.

The only accessible point for acquiring data was over the input shaft. The normal velocity
vibration spectra and acceleration time waveform acquired on April 14, 1997 are
presented in Figure 7. The speed reduction (input) gear mesh, 351 Hz, is dominant in the
spectral data in Figure 7 and showing significant side banding (especially at 2 x GM).
This pattern is indicative of gear wear and perhaps some misalignments. The P-P
acceleration data in time waveform was less than 4 g's and not considered significant.
This possible gear wear misalignment had been flagged with an action item to initiate a
visual inspection at next opportunity.

It was decided to apply the PeakVue methodology on April 14, 1997. The PeakVue
spectral and waveform data are presented in Figure 8. The only activity in the spectral
data is the output shaft turning speed with many harmonics. The time waveform has two
impacting regions per turn of the output shaft. The impacting levels exceed 40 g's. This
signature indicates a gear with significant cracked teeth (at root) in two regions. One of
the output shafts has the bull gear with 158 teeth and the pinion gear with 90 teeth driving

334
the second output shaft. The bandwidth in Figure 8 is not sufficient to encompass either
gear mesh; therefore we cannot identify from this data set which gear set has the
defective gear.!

From the spectral presented in Figure 7, one would be suspicious that the defective gear
would be in the GM I set since nothing unusual is present relative to the GM 2 set. There
was an additional PeakVue spectra taken with a bandwidth of 5000 Hz. In this PeakVue
spectra set, the GM 2 activity was present and the GM 1 absent. This leads to the
conclusion that the gears with the cracked teeth were most probably in the GM 2 set.

0,04 SM1-= TB" ROUTSPECTRM

-- 0.03 PK..0-08
_ ILOA80- tLO-0

So~2.
2.0 - A

S1.0 RM

K{-)
.-
- 1-8
I6
'S 0•

o CRESTF0
5.0
Poo*#*"s

1A. 1.0 200 FOT: 351.25

ooo~o~oOrdrWV
Time n mS• S.-: 210.0
.02..9

Figure 7. Velocity spectra and acceleration time wave form from the precision tension
bridle gearbox.

SM -Pýl1o= 8flonBrdWe

~ I
P
ANALYZESSF0000
144p-47 08:8:16

E
i I (Ft.O-
HP W00z)
RMS- .8
LOAD100.0
0RPM= 216.
RPS- 3.$0

S o lOO 1.o 2NO

F-.uecy ikHz

SRMS
SPPK(.) 7.16
=36.".
PK(.-) A
,2
SUCRKTF- .10
io

1.5 2.0 8.20

1F3O
TimeinS. d4 Spec: 3288

Figure 8. PeakVue spectra and time waveform from the precision tension bridle gearbox.

*The gear mesh where the defective gear is located would be present in the PeakVue spectra.

335
Following the acquisition of the PeakVue data, the gearbox was shortly shutdown and
inspected. One of the gears in the GM 2 set was found to have two visible cracks.

3.5 Extruder Gearbox: The plan view of the extruder gearbox is presented in Figure 9.
For the indicated input speed of 1840 RPM, the output shafts are turning at 316 RPM.
The gear mesh frequencies are 1318 Hz, 659 Hz, 237 Hz, and 79 Hz. For a gearbox of
this complexity, experience has shown a measurement point should be located at each
bearing.

The monitoring of the gearbox using normal vibration and PeakVue methodologies
identified worn gear sets, probable gear misalignment, defective bearings, and excessive
driver (DC Motor) speed variations. The signatures identifying the driver speed variation
are presented below.

32 T.

Output -

316 RPM

Output II)

32T.
3 148RPM -

GMr237 Hp~

CII (D ID N

H
F-GM-=659-
EN H 276 RPM

__________________
are prsneCnFgr
approximatelyC
1825 RPM 0 iC, h Omiatatvtstega
1840RPM tinput

Figure 9. Plan view of the extruder gearbox.

The velocity spectra and acceleration time waveform for an input shaft speed of
approximately 1825 RPM are presented in Figure 10. The dominant activity is the gear
mesh for the 23T/96T set and the 43T/62T set. There are many harmonics of the 23T/96T
set with the third being the largest (indicating looseness). The 43T/62T have reasonable
shaft speed side banding at 2 x GM, which suggests some misalignment.

The PeakVue spectra and time waveform for inlet shaft speed of 1833 RPM are presented
in Figure 11. There are activity at (1) the inlet shaft speed of 21.2 Hz (2) the first
.intermediate shaft speed at 31 Hz, and (3) the 43T/62T GM frequency and 2 times the
43T/62T GM frequency. The 2 times 43T/62T GM frequency is side banded with the
shaft speed on which the bull gear is mounted. The impacting at 2 times gear mesh is
indicative of significant back lashing which could be introduced with torsional resonance
or (more probable) significant inlet shaft speed fluctuation.

336
 Z1-Z1 N Extruder Gearbox
ZIN Gbox--V11 Vert, GR Mesh, Shaft 1
.16 1 1 1 1 ROUTE SPECTRUM
04-JUN-96 08:27
.14 (SST-Corrected)
) .12 OVRALL= .2033V-AN
"2PK6= .2061
LOAD=100.0
10 RPM= 1218.
I.08 RPS=20.31
.2 .06 1314 V628.6

04
Q .02 T39.9
0 1000 2000 3000 4000 5000

1i 1
Frequency
1[1
in Hz WAVEFORMDISPLAY
4

3. S3. I i = ' =: l ! 04-JUN-96

RMS= 1.21 08:27

2. PK(-)I::3.11
(b
'• 1.CRESTF 2.84

0 -2.
< -3.

-4. •Freq: 235.18

0 50 100 150 200 250 300 350 400 Ordr: 11.58

Time in mSecs Spec: .01871

Figure 10. Velocity spectra and acceleration time waveform from extruder gearbox on
June 4, 1996.

ZI1-Z1 N Extruder Gearbox

Z1 N Gbox-V1 1Vert, GR Mesh, Shaft 1
0 1 RPM"
108
-JL1272
M I I I ANALYZE
04-JN-6081SPECTRUM
S 0 .07 - RMS = . .005
I4JUN-9808:19
"--0RPM = 1271.
RPS =21.19
S0.06 1834RPM .

A 0.04

0.02
00<
ci)
0
0 600 1200 1800 2400 3000
Frequency in Hz

1.2
1.00
1.4

:
2 WAVEFORM
RMS
DISPLAY
84-JUN-9608:19
-. 2150
PK(.) =1.25

0.6
0.8
:1 ii ,.I i i
jPK(-
5.82
CRESTF
.4039

-0.4

-0.6
F06q 1313.7
0 60 120 180 240 300 Ordr: 61.99
Spec: .01069
Time in mSecs
Figure 11. PeakVue spectra and time waveform form extruder gearbox on June 4, 1996.

337
 Zl-Z1N Extruder Gearbox
ZlN Gbox-PTS =V1I V12

006
000

S004
003

002

1727RPý

0 600 1200 1800 2400 3000

Frequency in Hz

Figure 12. PeakVue spectra on extruder gearbox before DC motor speed adjustment
(1843 RPM) and PeakVue spectra on extruder gearbox after DC motor speed
adjustment (1727 RPM).

Spectra data on the inboard of the DC motor showed excessive* activity of 0.2 ips-peak at
the SCR frequency (360 Hz) with amplitude variation (side banding) at the motor shaft
frequency (inlet shaft to gearbox) and first intermediate shaft of the gearbox. The speed
controller was adjusted and measurements on the gearbox repeated. The velocity spectra
did not change, i.e., the 23T/96T set still showed signs of looseness and the 43T/62T set
still showed probable misalignment. There were significant differences in the PeakVue
spectra as shown in Figure 12. The spectra captured after adjustment of the speed
controller (1727 RPM) shows no indication of impacting.

3.4 Crusher Gearbox: This gear box, driven by a 8-pole 2000 HP motor, drives a rock
crusher at a mining facility. A plan view of the gearbox is presented in Figure 13. The
gearbox is nominally 17' x 12' x 7' in size. Normal vibration and PeakVue data was
acquired on a scheduled basis on the motor and, as much as possible, at each bearing with
the gearbox. The input shaft speed was in the proximity of 894 RPM. The first
intermediate shaft was turning at 531.5 RPM or 8.86 Hz.

The PeakVue spectra and time waveform data taken at measurement point 3 (see
Figure 13) are presented in Figure 14. The dominant activity in the PeakVue spectra are
the intermediate shaft turning speed (8.86 Hz) and many harmonics with the 4' , 8 th, 12th
etc. being dominant. In the PeakVue time waveform, Figure 14, there are four distinct
impacts per turn of the first intermediate shaft (the vertical lines are spaced at time
increments corresponding to the first intermediate shaft turning speed).

Vibration exceeding 0.1 ips-peak at the SCR firing frequency on inboard of a DC Motor generally imply a
problem in the controller circuit.

338
 Brg47 Brg 6 S@ev @r
U/PShaft I••

893 rpm

FiO1Pcshear
8 rp m
1rpmýi0
IS1 R5 3 1
Brg rp 10
Brg 3J
diatenterg7 Sleev r

interimrS.ti Brg 7 leeve Br

shaft

1~II:
Figure
0.2L- 13. Plan view of crusher gearbox. 55.

S0.18 28-Feb•l 10:30:14

(FWn44P
1000H.)
0.15
S~ OVERALL-W549
RM$=.3890
A-AN

0.09
...It... o 12

! LOAD

...
= 1o.o
RPM:=535.
RPS" .32

. 0.05 "
1I.~Li~tt~TrtT•..
.- LIi-T,
0.039.•

o so ....1oo 1so
t F~ [y 20 so 30 50o 40o
4:WAVEFORMODspLUy

3
, ~28,,Feb-97
P11"
RMN-A47O
M10:30:14

PK(- =;.22

0.2 0A S o. 3.0 1.2 Frq: .0

Time0
inSeconds Spec: .00178

Figure 14. PeakVue spectra and time waveform taken on crusher gearbox at point 3.

The first intermediate shaft has a beveled bull gear with 37 T and a pinion gear with 22 T
(see Figure 13). The first postulate was that the 24 T pinion gear had some fault at every
6 'h tooth since the total number of teeth (24) was divisible by 4. The bothersome fact with
this postulate was the g levels of the impacts were greater at measurement point 3 than at
measurement point 5 (the 24 T pinion was closer to measurement point 5). The impacts
were clearly occurring at four equal intervals per rev of the first intermediate shaft and
hence the 37 T bull gear was not considered to be the source where the impacting had
occurred (37 not divisible by four).

339
The gearbox was disassembled and inspected. The 24 T pinion showed no indication of a
problem. The impacting was occurring between the 37 T bull gear and the first
intermediate shaft. An approximate 2 in. band of fretting was completely around the
intermediate shaft at the top of the beveled bull gear.

The postulate then was the impacting was being introduced by a reasonable sharp
(high Q) torsional resonance of the input shaft. Strain gauges were installed near the
gearbox on the inlet shaft and torsional vibrations data acquired. The torsional resonance
spectra showed dominant activity at 35.3 Hz, which is 4 times the intermediate shaft
speed.

4. Conclusions: The capture and analysis of stress waves, which accompany many
classes of faults experienced in gearboxes, has proven to be an effective diagnostic tool
for fault detection and severity assessment in gearboxes. In this paper, five typical
examples of faults within gearboxes were presented as case studies. In each case, the
normal vibration analysis contributed very little to the fault detection and severity
assessment.

The PeakVue methodology for the capture and analysis of the stress waves provide a very
powerful trending capability. This is the case since the true amplitude of the specific
faults in g-units is captured independent of the machine speed, the analysis bandwidth
etc. This ability to capture the true impacting levels provides the knowledge to develop
absolute levels from which alert levels and alarms can be set based on a broad case
history library. Experience has shown these levels to be dependent on machine speed (in
a predictable manner) and fault type (the same as in normal vibration analysis).

340