Académique Documents
Professionnel Documents
Culture Documents
Army
Technical
Memorandum
107145
Technical
Research
Report
Laboratory
ARL-TR-970
Detecting
Gear Tooth Fatigue Cracks
in Advance
of Complete
Fracture
James
Lewis
J. Zakrajsek
Research
Center
Cleveland,
Ohio
and
David
G. Lewicki
Vehicle
Propulsion
Directorate
Prepared
1996
Ohio
for the
Technology
Showcase
Diagnostics
and Failure
sponsored
by the Society
Mobile,
Alabama,
April
on Integrated
Monitoring,
Prevention
for Machinery
22-26,
Failure
Prevention
Technology
1996
U.S. ARMY
National
Aeronautics
Space Administration
and
RESEARCH
LABORATORY
DETECTING
GEAR
TOOTH
FATIGUE
COMPLETE
James
National
CRACKS
IN ADVANCE
OF
FRACTURE
J. Zakrajsek
Aeronautics
and Space Administration
Lewis Research
Center
Cleveland,
Ohio
44135
and
David
U.S. Army
Lewis
G. Lewicki
Research
Research
Cleveland,
Abstract:
Results
presented.
bending
of using
An experimental
fatigue.
predictions
of fatigue
signal
tion research.
The damage
tooth
applied
from
fracture
propagation
consisted
failed
purpose
gear tooth
of these
rim fracture
through
both on-line
methods
positive
to respond
to the fatigue
and recorded
Key
Words:
stages
Gear;
Introduction:
propulsion.
of crack
Drive
The need
Fatigue;
notch
in
analytical
fault detection
methods
completed.
The objec-
prior to complete
rim fracture
rim gear tests. In the standard full rim gear test all of the methods responded
in advance of tooth fracture; however,
only three of the methods responded
in the early
through
a small
of detecting
was given.
crack
are
capable
detection
cracks
signal
fatigue
of complete
on the standard
to the vibration
methods
to detect
direction
a total of three
crack
crack
The vibration
single
methods
44135
were
vibration-based
Ohio
Laboratory
Center
in the thin
propagation.
Diagnostics;
train diagnostics
for a reliable
Failure
prediction
and usage
significant
monitoring
areas
system
of research
in rotorcraft
system
This would
plished
require
with the
a thirtyfold
aid of a highly
A key performance
as possible.
increase
element
Research
in helicopter
reliable,
of a HUM
on-line
system
is thus needed
safety.
Health
Practically,
and Usage
is to determine
to develop
Monitoring
if a fault exists,
and prove
various
fault
(HUM)
system.
detection
concepts
and
methodologies.
For rotorcraft
transmissions,
of gear tooth
fracture
damage.
system.
tion techniques
length
A number
of methods
signal.
A variation
using
the energy
non-linear
of the energy
associated
vibration
mance
of the detection
which
resulted
compared,
and overall
the fatigue
cracks
Theory
of Fault
in complete
time synchronously
tooth
resolution
Methods:
FM4
was developed
number
of teeth
ponents
(shaft
to detect
frequency
order
sidebands)
ment
(normalized
kurtosis)
[5] pro-
to detect
operator
detection
damage
to track
methods
performance
gear tooth
time
the
with
is an integral
damage.
to determine
the relative
data from
perfor-
Results
regarding
of each
gear
the performance
vibration
fault
detection
method
of the methods
in this investigation
data was converted
and vibration
incoherent
revolutions
are
in detecting
methods
and harmonics,
pattern
is first constructed
primary
meshing
the time-averaged
signal.
is then applied
to this difference
2
domain.
discussed
in the vibration
signal
by linear
utilized
to digital
vibration
form
and
of revolu-
interpolation
to provide
changes
from
The
noise
to the diagnostic
[2]. A difference
first
of these
relative
their
McFadden
of the demodulated
the energy
to experimental
to eliminate
when converting
using
of this research
fracture.
are made
as it was collected.
averaged
1024 samples
frequency
phase
Method
detec-
in which
of total fracture.
Detection
Methods
to detect
conclusions
in advance
tests,
on gear teeth.
of their
means
the objective
The vibration
The primary
damage
Verification
gear.
with a comparison
it becomes
methods
general
by Swansson
of an accurate
of the
gear fault
fatigue
on gear teeth.
Ma [7] proposed
a damaged
along
tooth
failure
by applying
detection
spectrum.
development
damage
ratio reported
with
data
early
isolated
cracks
into complete
catastrophic
propagation
detection
gages.
[3,4] to detect
gear tooth
in the frequency
energy
experimental
Lewis
to detect
crack
to provide
[2] to detect
is the accurate
can propagate
of gear failure
spur gear
crack
system
in a potentially
this mode
special
HUM
if undetected,
can result
several
at NASA
a method
addresses
using
of a reliable
crack,
of which
data from
was monitored
developed
posed
This paper
element
fatigue
both
to vibration
A gear tooth
transmission
crack
a critical
This averaged
the
and resampled
below.
resulting
by removing
from damage
the regular
frequency
and harmonics
The fourth
normalized
signal.
on a limited
meshing
along
statistical
comwith
mothe
time signal
is the Hilbert
signal
and
sidebands.
is then created
transform
represents
Amplitude
with a value
Another
demodulation
of the signal.
an estimate
modulation
variations
in the loading.
transient load fluctuations
sionless,
in which
envelope
of the amplitude
in a signal
signal,
is the magnitude
modulation
is most often
present
of this complex
in the signal
due to periodically
reoccurring
due
time
to the
transient
of 3 under
technique
nominal
was
conditions.
developed
to detect
local
gear
defects,
such as gear
tooth
revolution
method
in order
to quantify
the relative
variance
of
in the run.
investigated
is a variation
to be a robust
of the
indicator
energy
originally
constructed
dividing
the standard
deviation
of the difference
of heavy
meshing
components,
as defined
previously.
ratio
uniform
signal
In this study,
3
technique
[6]. The
energy
by the standard
the energy
deviation
ratio is found
ratio
ratio is found
was
by
of the regular
by dividing
the
sum of thenon-mesh
related
amplitudes
frequencies
trend
(meshing
amplitudes
the damage
in the frequency
spectrum
spectrum
of increasing
This parameter
amplitudes
related
at non-mesh
related
the
frequencies
as
developed
for
increased.
A technique
using
the "energy
operator"
was
speech production
modeling,
the energy
recently by Ma [7]. The energy operator
a vibration
signal.
An impulse
tooth is reinforced
this study,
in the time-averaged
by the energy
the time
averaged
quency
using
applied
to the band-passed
operator,
signal
+20 sidebands.
also investigated
The
signal.
in this study.
First
operator
was recently
proposed
for gear defect
is a non-linear
function that reacts to impulsive
vibration
thus allowing
the impulse
is first band-passed
discrete
form
In order
to detect
signal
filtered
of the energy
caused
the primary
operator,
any reinforced
by a damaged
to be more easily
about
as given
impulsive
detection
energy in
gear
detected.
In
meshing
fre-
in Ref. 7 is then
events
in the energy
operator,
the normalized
kurtosis is then applied to the energy operator results. The normalized
kurtosis, as described
earlier, reduces the results to one nondimensional
number with the value of 3
under nominal conditions.
Apparatus
and Gear
Damage
Review:
shown
in Figs.
of crack propagation
tests conducted
on a spur gear fatigue test rig at NASA Lewis Research
Center. To initiate cracks, a notch was placed in the fillet region (loaded side) on one tooth of each
of the test gears.
the position
specimen
The notches
of the maximum
through
predictions
To verify
were located
tensile
bending
of fatigue
the crack
fatigue.
crack
stress.
The
propagation
growth
at a radius
rate, crack
of 40.49 mm (1.594
is
The spur gear test rig was then used to fail the spur gear
primary
purpose
direction
detection
of these
tests
were installed
was to verify
analytical
[8].
with
was captured
using
an on-line
program
running
on a personal
computer
with an analog
to
digital conversion
board and anti-aliasing
filter. The 28-tooth instrumented
test spur gear meshes
with a 28-tooth standard rim spur gear at a nominal speed of 10 000 rpm in the fatigue test rig. The
spur gears have a face width of 6.35 mm (0.25 in.) and a pitch diameter of 88.9 mm (3.50 in.). Each
test started with a one hour break-in period at a low load level, after which the load was increased
and the test continued
for runs
until complete
fracture
Runs
crack
gages,
period,
applied.
length
is plotted
as a function
started
to failure.
For runs
(1200
illustrated
in-lbs)
problems
was used
was used.
in Fig. 1. Crack
length,
as recorded
by the
of run time for run 1 in Fig. 3. The first hour was the break-in
the crack
1 and 2, failure
lines connecting
Due to instrumentation
in Fig. 1. The
length
occurred.
gages
cover
is defined
as the complete
gage
gages
approximately
points
to failure
plot is available.
1.75 hours
is for reference
correctly
purposes
only.
total crack
length
illustrated
4. The break-in
to failure.
applied.
in figure
period
2. Crack
is not included
is defined
length
is plotted
gages
cover
as complete
senting
a moderate
crack
time represents
crack
is achieved.
length
Discussion
of Results:
Results
signal
period.
9 illustrates
point)
from
rate
(first 75%
a region
in Figs.
time-averaged
Figure
growth
of propagation
of accelerated
of applying
crack
the diagnostic
5, 6, and 7, respectively.
time).
growth
of the tooth,
to crack propagation
8 gives
the frequency
the time-averaged
vibration
plots
runs
spectrum
1, 2,
of the
as seen in
rate, where
methods
Figure
of
started
fracture
figure 2. Although
the crack gages only cover 18% of the total crack length,
time from the start of the crack to failure. Two regions can be seen in figure
propagation
as a function
point
(end
of test
revolution
of
on the results,
fatigue
crack
the vibration
plotted
the gear
in Fig.
(Fig.
5(b))
detect
were
unable
to give early
parameters
Each
and NB4
failure.
5, with
the crack
pending
methods
data
as a function
seen
fault detection
the exception
(Fig.
5(c))
until complete
The kurtosis
of the kurtosis
reacted
of the energy
as a function
of the energy
to the damage.
rim fracture
period.
resulted,
However,
operator
appears
of the
were applied
of crack length
NA4
no advanced
are
also. As
only parameters
to have reacted
to
of the parameters
operator,
thus giving
indication
NA4
unable
warning
to
of the im-
in the
run, as seen in Fig. 5(f). This very large reaction at the start of crack propagation
is not repeated
in
runs 2 and 3. It is unclear at this time as to why this ideal type of reaction occurred
only on run 1.
Figure
6 plots
propagation
detection
the results
of the parameters
plot is available
methods
give positive
detection
indication
methods
3. Each
of the parameters
length
(top of plot).
As seen
region
of accelerated
crack
crack
length
nondimensional
is achieved
parameters,
As stated
in only
25%
NB4 peaks
in run 3, more
advanced
of crack
propagation.
as a function
earlier
crack prior
of the
warning
at a value
of the crack
during
trend
of establishing
above
tooth
period
of crack
to the crack
in the
where
time.
Of the
of the energy
normalized,
operator
are referenced
the nominal
operator
value
peaks
to the
propagation
catastrophic
of the energy
an exceedance
fault
fracture
1. The
and as a function
substantially
growth
no crack
of the test.
to complete
is needed
an increasing
A method
crack
the crack
that show
react
total
seen in run
the initial
of run time,
(Fig. 7(b)),
to the fatigue
to those
are plotted
are similar
of the damage
reacted
during
event.
As
(Fig. 7(f))
the first
relative warningtimesfor the threeparameters.This methodusesa valueof five standarddeviations over the meanvaluefor the exceedance
level, wherethe meanlevel andstandarddeviation
arebasedonthe dataprior to crackgrowth[9]. As seenin Fig. 7(f), basedonthis exceedance
level,
thekurtosisof theenergyoperatorgivesthe earliestwarningof damage.It reachesthe exceedance
level at 0.083hours,or at 46% of the total crack propagationtime, at which the crack hasonly
advancedto approximately7%of its total length.As seenin Figs.7(b) and(c),NA4 andNB4 reach
the exceedance
level at 0.111hours,or at 62%of total crackpropagationtime, atwhich the crack
hasonly advancedto approximately11%of its total length.Thewarningtimesbasedon thesethree
parametersare nearlytwice aslong asthatgivenby the otherfault detectionmethods.In addition,
NA4, NB4, andkurtosis of the energyoperatorwere able to detectthe crack during the more
moderatecrackgrowthrateregion.
The additionalcomplianceassociatedwith the thin rim gearsalongwith the geometryof
gears
plots
the frequency
the parameters
spectrum
from detecting
the fatigue
point
after
crack
in runs
the break-in
these
1 and 2. Figure
period
for each
run.
The
spectrums
from the thin rim gear tests, Figs. 8(a) and (b), show a more complex frequency
distribution as compared
to the spectrum
from the full rim gear test (Fig. 8(c)). As seen in Fig. 8(c), the
frequency
harmonics
spectrum
is primarily
composed
of the primary
gear mesh frequency
of the mesh frequencies,
and the corresponding
sidebands
about these
non-uniform
thin rim
geometry
illustrated
in the plots
shown
complex
meshing
pattern
rim fracture
has occurred
is clearly visible
are nonlocalized
a more
has occurred
in this figure
associated
a standard
with the
rim gear.
compliance
than
induced
impulse
tooth to create
the impulse
in the vibration
signal,
thus compensates
for the fault until complete fracture.
In addition, the nonuniform
geometry
thin rim gear complicates
the vibration
signal, possibly
masking the dynamics
of the tooth
Conclusions:
Based
perimental
on the results
data from
indication
of applying
of the fatigue
a variety
propagation
ated with
the thin-rim
gears
influence
of the crack
on the vibration
may have
prior
NA4,
NB4,
detection
techniques
The
detection
and
of the
crack.
techniques
to ex-
conclusions
can be made.
investigated
were unable
additional
compliance
thus minimizing
to
associ-
the dynamic
reacted
to the fatigue
crack,
in
tooth fracture.
The more robust reactions
to the fatigue crack
propagation,
as the crack accelerated
to complete
fracture.
was growing
the following
to rim fracture.
(2) In the full rim spur gear test, all of the techniques
varying
occurred
of gear fault
tests,
compensated
signal
for
operator
techniques
at a moderate
rate.
reacted
reactions
crack
to the fatigue
during
the early
crack
stages,
References:
1. Astridge,
2.
D.G.:
Helicopter
Transmissions
- Design
for Safety
and Reliability.
Proc., Pt. G-J Aerosp. Eng. vol. 203, no. G2, 1989, pp. 123-138.
Stewart, R.M.: Some Useful Data Analysis Techniques
for Gearbox
R/10/77,
Machine
versity
Health
of Southampton,
3. Zakrajsek,
to Pitting
Failures
Prevention
to Spiral
Bevel
Prevention
Data.
Office
of Sound
of Naval
Gear Fatigue
Office
Vibration.
Proceedings
Report
and Vibration
MHM/
Research,
Fatigue
Research,
Cracks
J. Vib. Acoust.
Arlington,
H.J.: Application
Data. Proceedings
of Naval
of the 47th
Research,
P.D.: Detecting
the Meshing
Institute
H.J.; An Analysis
Failure
Group.
Group.
5. McFadden,
D.P.; Decker,
Fatigue
Handschuh,
J.J.;
Group,
Diagnostics.
Eng.
Uni-
July 1977.
Townsend,
J.J.;
Applied
4. Zakrajsek,
Monitoring
Inst. Mech.
Meeting
in Gears
Stress
Techniques
of the Mechanical
Failures
by Amplitude
Reliab.
as
Methods
of the Mechanical
and Phase
Design,
vol.
Demodulation
of
1986,
pp.165-170.
6.
Swansson,
Conference
Australia,
7.
N.S.: Application
of Vibration
Signal Analysis
Techniques
to Signal Monitoring.
on Friction and Wear in Engineering
1980, Institute of Engineers,
Australia.
Barton,
1980, pp. 262-267.
8. Lewicki,
Institute,
D.G.:
for Aerospace
9.
Rose,
Operator
and Other
Crack
Fatigue
Failure.
Office
of Naval
Illinois,
propagation
Thin-Rim
H.J.: Vibration
Gears.
Signature
Proceedings
Research,
Demodulation
of the Society
Approaches
for Machinery
Prevention
Detection.
Pro-
Te c h no lo g y.
Studies
to Determine
Ph.D. Dissertation,
and Fatigue
Virginia,
Benign
Case Western
Crack
to Gear Defect
Failure
Growth
or Catastrophic
Reserve
Analysis
of the Mechanical
1990.
Failure
University,
May
of a Gear Tooth
Failure
Modes
Prevention
1995.
Bending
Group.
C-94-02303
Figure 1 .--Gear
geometry
configuration
C-94-02304
Figure 2.--Gear
100-
o Front side
f-
80 -
[] Rear side
x Complete rim
fracture
60-
100
;_
i
// /
/
/
/
/
/
/
o o_
//
ID
Q.
I
2
I
4
60
o
40
x Complete
fracture
tooth
i//I
Ii
I
-
i
/
I
(1)
Q.
I
8
I
6
as a function
80
0
0.00
0.04
0.08
0.12
L
0.16
Run time, hr
Run time, hr
Figure 3.--Crack
length
run time for run 1.
._ __
//
200
[] Rear side
03
for run 3.
o Front side
t--
configuration
03
geometry
of
Figure 4.--Crack
length
run time for run 3.
as a function
of
Crack length
Crack length
(percent
of total
front
length
(percent
to fracture,
of total
front
length
to fracture,
50
50
10-
4
_E
LL.
22
1-
(a)
I
2
0o
I
4
I
6
(b)
I
8
0o
i
2
50
50
1_
3_
oo(C)
5_
Run time, hr
Run time, hr
50-
21F33[;00
4030,
m
2010-
I
2
I
4
I
6
Run time, hr
Run time, hr
50
50
0.25
FIIF
I
0.20
o.e 100_,
80-
O
,B
0.15
O.lO
E
0.05
I
8
(e) i
0.00
0
I
2
I
4
I
6
I
8
Run time,
Run time, hr
Figure 5.mResponse
parameters
hr
for run 1.
tn
tO
i
i
!--
lb.
o._
r"
drr
c;rr
da:
=
E
I_1
I_
LO
LO
_"
03
04
I--
o.
I_
143
O4
Q.
C
O O
LO
Lf)
.g
O4
eseqd snoeuelue_su!
JO:_eJedo _6JaUe
jo _teed-o;-_leed
tTVN
jo s!so_Jn_l
to
tO
I
O
I-
L
T-
E
r-
-drr
-dn"
O
)O
t7_-4
I_l
O3 X) I_
_SN
od
14") O
_O
c_ d
_f)
o!;e_/[6Jeu3
c_ c_ o
Crack
(percent
of total
length
Crack
length
to fracture,
front
11
8-
(percent
gage)
of total
length
length
to fracture,
11
15
3.55.88.111Ll13/18100
1 1
3.5
1
18
front
gage)
15
5.88.1
1
1111311181001
16
14
12
4
*d"
LL
10
"
_+5_
2-
1 -
0 (a)
0.00
I
0.04
I
0.08
I
0.12
120
0.04
0.08
0.12
hr
Run time, hr
1115
1115
Run time,
3.5
1
2
0
0.00
I
0.16
5.88.1
1
I
/18
[113 I I
1001
0.16
15
(9
9
100
N--
o {10
8O
m
z
60-
6_C
40 _
0.00
_+5_r
0.04
0.08
0.12
.c_ 0 (d)
0.00
0.16
I
0.04
I
0.08
1115
1115
0.30-
I
0.16
Run time, hr
Run time, hr
3.55.88.11113[18100
I
0.12
=-
1 1
11111
100
f-0.25
X+5_
o 0.20
(9
e,(9
>, 0.15
(9
_- 0.10
LU
.__
0.05
o.oo (e)
0.00
I
0.04
t = .1444 -_
I
I
0.08
0.12
ffi
0
0.16
_"
0.00
0.04
Run time, hr
Figure 7.--Response
0.08
0.12
Run time, hr
of gear fault detection
11
parameters
for run 3.
0.16
-n
bo
Vibration
amplitude
Vibration
amplitude
Vibration
amplitude
REPORT
Public
reporting
gathedng
collection
Davis
burden
for
and maintaining
of information,
Highway,
Suite
this
DOCUMENTATION
collection
of information
is estimated
the data
needed,
and completing
including
suggestions
for reducing
1204,
Arlington,
VA
22202-4302,
and
to average
and reviewing
this burden,
to
to
the
Office
1 hour
per
the collection
Washington
of
Form Approved
OMB No. 0704-0188
PAGE
response,
Management
and
the
Send
Services,
Budget,
2. REPORT DATE
January
including
of information.
Headquarters
time
for
comments
Directorate
Paperwork
reviewing
Gear
searching
existing
data
sources,
Reduction
Project
(0704-0188),
Washington,
DC
20503.
Technical
instructions,
regarding
this burden
estimate
or any
other
aspect
of this
for
Information
Operations
and Reports,
1215
Jefferson
Memorandum
5. FUNDING NUMBERS
Tooth
Fatigue
Cracks
in Advance
of Complete
Fracture
WU-505-62-36
6. AUTHOR(S)
James
1L162211A47A
J. Zakrajsek
and David
G. Lewicki
8. PERFORMING ORGANIZATION
REPORT NUMBER
E-10075
10. SPONSORING/MONITORING
AGENCY REPORT NUMBER
NASA
TM-107145
ARL-TR-970
Machinery
Failure
and David
G. Lewicki,
person,
Prevention
Vehicle
James J. Zakrajsek,
Subject
on Integrated
Mobile,
Propulsion
organization
12a. DISTRIBUTION/AVAILABILITY
Unclassified
Showcase
Technology,
Monitoring,
Alabama,
Directorate,
code 2730,
Diagnostics
April 22-26,
U.S.Army
and Failure
Prevention
Research
Laboratory,
NASA
sponsored
by the Society
Lewis Research
Center.
for
Center
Responsible
(216) 433-3968.
12b. DISTRIBUTION CODE
STATEMENT
- Unlimited
Category
This publication
37
Information,
(301) 621_)390.
of using
initiated
through
analytical
signal
based
rim
fault
objectives
fracture
failure
crack
were
were
detection
stages
crack
rim
tooth
bending
of a tooth
propagation
gears
given.
on the gear.
The
primary
Results
of tooth
for gear
tooth
fracture
signal
both
of detecting
showed
rim
fracture;
fault
on-line
however,
and after
crack,
detection
In the standard
only
three
Fatigue;
of these
tests
rim thickness.
The
in each
test
was
was
to verify
The
vibration
damage
consisted
full rim
test gear.
Vibration-
the tests
were
and determine
methods
full rim
of the methods
failed
completed.
how
The
far in advance
to respond
gear test
of
to the
responded
to the fatigue
propagation.
crack
research.
standard
the fatigue
of gear
detection
on the
An experimental
fatigue
purpose
and single
in the thin
are presented.
gear tooth
to the vibration
fracture
cracks
The
direction
capable
fatigue
fatigue.
and recorded
methods
in advance
of crack
area
applied
was
rim
gear
through
monitored
to identify
to complete
to the fatigue
in the early
was
two thin
methods
positive
prior
crack
tests
on the
to detect
specimens
in the fillet
of fatigue
detection
fatigue
notch
of three
of this effort
of total
responded
a small
a total
methods
of spur gear
predictions
from
complete
crack
vibration-based
to fail a number
Diagnostics;
Failure
prediction
Unclassified
Standard Form 298 (Rev. 2-89)
Prescribed by ANSI Std. Z39-18
298-102
Center
If Undeliverable--
Do Not Return