-
aus
ulo-265
VibrationCondition monitoringGearbox
f aocesenrimasiso exnito
highlighted. The differences in the parameters evolution of each
NDT technique are discussed and the
ains inan leas the inon-linechniquh effor
250 h) and correlated special features (based on higher order
mo-ments) extracted form the vibration recordings with the Fe
debrismass accumulated during the tests. They have integrated their
re-sults in a fuzzy logic based health monitoring system with
satisfac-tory performance. Researchers in the eld have focused
mainly onadvanced signal processing techniques applied on vibration
record-ings coming mainly from articial gear defects in short tests
ratherthan inducing gear pitting damage in multi-hour testing.
the power spectrum distribution. The same authors have also
ap-plied the WignerVille distribution [12] as well as the
wavelettransform [13] on vibration and acoustic signals for the
samepurpose.
The interest for applications of acoustic emission (AE) for
condi-tion monitoring in rotating machinery is relatively new and
hasgrown signicantly over the last decade. AE in rotating
machineryis dened as elastic waves generated by the interaction of
twomedia in motion, i.e. a pair of gears. Sources of AE in
rotatingmachinery include asperities contact, cyclic fatigue,
friction, mate-rial loss, cavitations, leakage, etc. AE technique
has drawn
* Corresponding author.
Applied Acoustics 70 (2009) 11481159
Contents lists availab
A
l seE-mail address: [email protected] (V.
Kostopoulos).Few research teams have published experimental data
coming fromlong-term testing to see the effect of natural gear
pitting mostlyupon vibration recordings. Dempsey et al. [14] and
Decker withLewicki [5] have conducted some excellent experimental
work atGRC/NASA and published interesting results from extensive
geartesting at a special test-rig utilizing vibration and oil
debris mea-surements. With the clear goal to improve the
performance of thecurrent helicopter gearbox health monitoring
systems, they havetested gears at high shaft speed for multi-hour
periods (up to
features of vibration signals and showed that unlike the
time-fre-quency distribution, which incorporates a constant time
and fre-quency resolution, the wavelet transform can
accommodatesimultaneously both the large and small scales in a
signal, en-abling the detection of both distributed and local
faults. Baydarand Ball [10,11] have proposed the instantaneous
power spectrumand have shown that it is capable in detecting local
tooth faults instandard industrial helical gearboxes. The
propagation of localfaults was identied by monitoring variations in
the features of1. Introduction
In gearboxes and power drive trdetection is often very critical
and caviation and in industry as well. Thunon-destructive
inspection and/orgrowing and effective diagnostic teare the
objective of extensive researc0003-682X/$ - see front matter 2009
Elsevier Ltd. Adoi:10.1016/j.apacoust.2009.04.007superiority of AE
over vibration recordings for the early diagnosis of natural wear
in gear systems isconcluded.
2009 Elsevier Ltd. All rights reserved.
general, gear damaged to increased safety interest for their
periodichealth monitoring ises and methodologiests over the last 50
years.
The publications in the eld of condition monitoring via
vibra-tions are quite versatile. Selecting a few and focusing on
advancedsignal processing techniques the works of Wang and
McFadden[6,7] must be mentioned, that utilized time-frequency
analysistechniques and showed that the spectrogram has advantages
overWignerVille distribution for the analysis of vibration signals
forthe early detection of damage in gears. The same authors
havealso employed the wavelet transform [8,9] to analyze the
localAdvanced signal processingAcoustic emission posed utilizing
the discrete wavelet transform. The evolution of selected
parameters/features versus
test time is provided, evaluated and the parameters with the
most interesting diagnostic behaviour areCondition monitoring of a
single-stage gecracks utilizing on-line vibration and aco
T.H. Loutas, G. Sotiriades, I. Kalaitzoglou, V.
KostopoDepartment of Mechanical Engineering and Aeronautics,
University of Patras, Patras GR
a r t i c l e i n f o
Article history:Received 7 October 2008Received in revised form
15 April 2009Accepted 16 April 2009Available online 17 May 2009
Keywords:
a b s t r a c t
The condition monitoring omethodologies and the prniques is the
aim of the prefor this purpose. The expepresented in detail.
Emphemission signals in order tnostic value from the mo
Applied
journal homepage: www.ell rights reserved.rbox with articially
induced geartic emission measurements
s *
00, Greece
lab-scale, single stage, gearbox using different non-destructive
inspectionssing of the acquired waveforms with advanced signal
processing tech-t work. Acoustic emission (AE) and vibration
measurements were utilizedental setup and the instrumentation of
each monitoring methodology areis given on the signal processing of
the acquired vibration and acoustictract conventional as well as
novel parametersfeatures of potential diag-red waveforms.
Innovative wavelet-based parametersfeatures are pro-
le at ScienceDirect
coustics
vier .com/locate /apacoust
-
attention as it offers some advantages over classical vibration
mon-itoring. First of all, as AE is a non-directional technique,
one AE sen-sor is sufcient in contrast to vibration monitoring
which mayrequire information from three axes. Since AE is produced
atmicroscopic level it is highly sensitive and offers
opportunitiesfor identifying defects at an earlier stage when
compared to othercondition monitoring techniques. As AE mainly
detects high-fre-quency elastic waves, it is not affected by
structural resonancesand typical mechanical background noise (under
20 kHz). Tandonand Mata [14] applied AE to spur gears in a gearbox
test-rig. Theysimulated pits of constant depth but variable size
and AE parame-ters such as energy, amplitude and counts were
monitored duringthe test. AE was proved superior over vibration
data on early detec-tion of small defects in gears. Singh et al.
[15] also applied AE tech-nique in condition monitoring of test-rig
gearboxes, whilevibration methods was also used for comparative
purposes byplacing accelerometers on the gearbox casing. They also
concludedthat AE provided early damage detection over vibration
monitor-ing. Toutountzakis et al. [16] investigated the inuence of
oil tem-perature and of the oil lm thickness on AE activity and on
AEsignals captured during continuous running of a
back-to-backgearbox test-rig. It was observed that the AE RMS
varied with timeas the gear box reached a stabilized temperature
and the variationin AE activity RMS could be as much as 33%.
Tan and Mba [17] discussed in more detail the oil
temperatureeffect on AE and concluded that the source of AE
mechanism thatproduced the gear mesh bursts was from asperities
contact. Tout-ountzakis and Mba [18] presented some interesting
observationson AE activity due to misalignment and natural pitting
and con-cluded that the AE technique is applicable for monitoring
geardamage. Finally a comparative study [19] between AE and
vibra-tions was conducted to show the diagnostic and prognostic
capa-bilities of each technique in several multi-day tests in a
single-stage gearbox.
The present work reports the results concluded by long term(50
h) experiments to a defected gear system, with a transversecut of
25% of root thickness to simulate the tooth crack.
Differentparameters, resulted by the analysis of the recording
signals (bothcoming from vibration monitoring and AE) are presented
andtheir diagnostic value is discussed in the direction of being
used
D
d
W
Table 1Conventional parameters calculated from the acquired
waveforms.
Time domain parameters Frequency domain parameters
p1 PN
n1xnN p12
PKk1skK
p2 PN
n1 xnp12
N1
rp13
PKk1 skp12
2
K1
p3 PN
n1jxnj
pN
2p14
PKk1 skp12
3
Kp13
p 3
p4 PN
n1 xn2
N
rp15
PKk1 skp12
4
Kp213
p5 max jxnj p16 PK
k1 fkskPKk1sk
p6 PN
n1xnp13
N1p32p17
PKk1fkp16
2skK
r
p7 PN
n1xnp14
N1p42p18
PKk1 f
2kskPK
k1sk
s
p8 p5p4 p19 PK
k1 f4kskPK
k1 f2ksk
s
p9 p5p3 p20 PK
k1 f2k skPK
k1skPK
k1 f4ksk
qp10 p4
1N
PNn1 jxnj
p21 p17p16p11 p5
1N
PNn1 jxnj
p22 PK
k1 fkp163sk
Kp317
p23 PK
k1 fkp164sk
Kp417
p24 PK
k1 fkp161=2sk
Kp17
p
T.H. Loutas et al. / Applied Acoustics 70 (2009) 11481159
1149Fig. 1. Test bench setup.
Time synchronous
averaging (only for vibration
signals)
Vibration and AE signalsFig. 2. Flow chart of the DWiscrete
Wavelet transform n levels of
ecomposition
Energy content determination for each level
Plot of energy levels vs
defect types
avelet type
Number of levels T-based methodology.
-
for the development of a condition monitoring system.
Further-more, a systematic comparison of the different diagnostic
param-eters is provided, in order to assess which are the most
robustand reliable ones for the condition monitoring of gearboxes
anddrive trains. The paper closes with the conclusions drawn
fromthis study.
2. Experimental setup
Fig. 1 shows the experimental setup used for the gears
testing.The test-rig consists of two gears made from 045M15 steel
with amodule of 3 mm, pressure angle 20, which have 53 and 25
teethwith 7 mm face width. The axes of the gears are supported
bytwo ball bearings each. The entire system is settled in an oil
basinin order to ensure proper lubrication. The gear box is powered
by amotor and consumes its power on a generator. Their
characteristicsare as follows:
1 stage gearbox with two gears (25 and 53 teeth); 3-phase 5 hp
motor (220 V, 9 A, 50 Hz, 1400 rpm) con-
trolled by inverter; single phase generator with continuous
power consump-
tion control (load uctuation), 4.2 KVA, 3000 rpm, 50 Hz; the oil
pump is of the wet type without oil recirculation; the shafts are
ball bearing supported.
Two non-destructive techniques have been employed to moni-tor
the gearbox during operation, namely vibration and
acousticemission. Two Bruel & Kjaer accelerometers were used
for thevibration monitoring both mounted upon the gearbox case,
onein each side-axis. The sampling frequency used was 50 kHz
andsignals of 1 s duration were recorded. Additionally three wide
bandFig. 3. Tooth crack.
0 25 300 350 400 450 500 550 600 650
0.0
0.2
0.4
0.6
0.8
1.0 (tooth break)
700%
1500%
2nd transition
3rd transition
1st transition
ED1
# of recordings
0.4
0.6
0.8
1.0
aram
ete
r
0.4
0.6
0.8
1.0
par
ame
ter
(a) (b)
(c)
1150 T.H. Loutas et al. / Applied Acoustics 70 (2009) 114811590
25 400 450 500 550 600 650
0.0
0.2 450%
2800%
p7# of recordings
Fig. 4. Parameters evolution during the test for vib0 25 400 450
500 550 600 650
0.0
0.2350%
1100%
p6 p
# of recordings
0 25 400 450 500 550 600 650
0.0
0.2
0.4
0.6
0.8
1.0
300%
90%
p13
para
mete
r
(d) # of recordings
ration ch1 (a) ED1, (b) p6, (c) p7 and (d) p13.
-
acoustic emission sensors manufactured by Physical Acoustics
Cor-poration (PAC) with a frequency response range of 100800
kHzrecorded continuous AE signals of 100 ms duration at a
samplingrate of 2 MHz. Fig. 1 shows the positions of all the
sensors.
One AE sensor is mounted on the output shaft (AE channel 1),the
second is placed upon one of the bearings of the same shaft(AE
channel 3) and the third (AE channel 2) is in friction contactwith
the input rotating gear. A special innovative device was de-signed
in-house and discussed elsewhere [20] in order to mountthe AE
sensor upon a rotating component without the expensive/demanding
solution of the slip-ring generally used in literature.
The recordings of all the above data coming from accelerome-ters
and AE sensors are realized by a National Instruments NI-6070
1MS/SEC FIREWIRE data acquisition device and assisted byspecial
software developed in-house, in Labview programmingenvironment.
Finally, the temperature of the oil bath within thegearbox is
measured via a thermocouple.
3. Signal processing methodologies
Signicant effort was dedicated to the signal processing of
thevibration and AE waveforms acquired during the tests. The
goalset a priori was to calculate a number of parametersfeatures
ex-tracted by the signals and check their behavior during the
tests
in order to identify the most promising ones that may be usedfor
damage detection and condition monitoring of the gear system.In the
literature, very few research groups have been involved inlong term
gear testing and they have mainly used higher order mo-ments and
their combinations to form diagnostic parameters [15]with
interesting behavior during the tests. In this work, apart
fromparameters usually found in the literature, we have
introducedsome more advanced signal processing techniques such as
the dis-crete wavelet transform and extracted innovative
wavelet-basedparameters from the signals. In total more than 40
parametersare checked for their diagnostic ability. Those capable
of monitor-ing the damage are identied and compared.
3.1. Conventional parameters
In Table 1 conventional parameters from the time and fre-quency
domain that were calculated, are shown. Where x(n) is asignal
series for n = 1, 2, . . . , N, N is the number of signal
samplesand s(k) is the Fourier transform for k = 1, 2, . . . , K, K
is the numberof spectrum lines, fk is the frequency value of the
kth spectrum line.Parameter p1 is the mean of the signal, p2 is its
root mean square, p5is obviously the absolute maximum of the
signal, p4 the standarddeviation, p6 and p7 the third and fourth
moments whilst p8p11 re-sult as a combination of previous
parameters all calculated by the
0.006
0.008
0.010
0.012
0.014
25%
1st transition
ED1
0.0005
0.0010
0.0015
0.0020
2200%
1st transition
p6 p
aram
ete
r
(a) (b)
T.H. Loutas et al. / Applied Acoustics 70 (2009) 11481159 11510
50 100 150 200 250 300 350 4000.000
0.002
0.004
# of recordings
0 50 100 150 200 250 300 350 4000.00000
0.00005
0.00010
0.00015
122%
1st transition
p7 p
aram
ete
r
(c) # of recordings
Fig. 5. A magnication of Fig. 2 o0 50 100 150 200 250 300 350
400
0.0000
# of recordings
0 50 100 150 200 250 300 350 4000.04
0.06
0.08
0.10
p13
para
mete
r
(d) # of recordings
ver the rst 420 recordings.
-
signal in time domain. Correspondingly, p12p24 are extracted
inthe frequency domain.
These parameters are typical parameters in the time and in
thefrequency domain that can be extracted from any signal.
3.2. Discrete wavelet transform based parameters
The wavelet transform was utilized to develop new parametersand
check their behavior during the tests. The major advantage
ofwavelets is their inherent ability to perform local analysis
withvarying precision. Wavelet transform treats low frequencies
withlow resolution and high frequencies with high resolution
[21].Wavelets stem from the iteration of lters and lter banks
(withrescaling) so they are inherently orthogonal or biorthogonal.
Incontrast to the Fourier analysis, which consists of breaking up
asignal into sine waves of various frequencies, wavelet
analysisbreaks up the signal into shifted and scaled versions of
the original(or mother) wavelet.
The inverse discrete wavelet transform can be expressed as:
f t cXj
Xk
DWj; kwj;kt 1
where c is a constant depending only on w. Eq. (1) is the
backboneof the present work and the whole philosophy of using
waveletsfor analysis of transient and non-stationary signals, as it
statesthat a given time series signal can be decomposed by the
discrete
wavelet transform into its wavelet levels, where the summation
ofthese levels represent the original input signal. The
decomposedwavelet levels are channeled in such a way that each
level corre-sponds to a certain frequency range of the acquired
signal. TheDWT-based methodology used in this work was introduced
anddescribed elsewhere [21,22]. Fig. 2 schematically summarizes
thecomplete procedure. It comprises the Discrete Wavelet
Transform(DWT) of the time synchronous averaged acquired vibration
sig-nals and AE signals in 10 levels of decomposition using thedb10
wavelet. As far as the type of wavelet for the discrete trans-form
is concerned db10 was a good compromise of smooth func-tion,
without sharp edges as in the case of db wavelets of
lowerorder.
The family of Daubechies wavelets was chosen because it
con-sists of biorthogonal, compactly supported wavelets,
satisfactorilyregular though not symmetrical. Other wavelets having
similarproperties to the Daubechies family, such as symlets or
coietswere also tried with minor impact upon the results. The
decom-posed wavelet levels are split in a way that each level
correspondsto a certain frequency range. After the 10-level
decomposition theenergy of each level (10 details and one
approximation) is calcu-lated. Thus eleven parameters namely
ED1ED10 (for the details)and Ea10 (for the approximation) were
resulted.
Additionally the sub-band wavelet entropy (SWE) is
calculated.SWE is dened in terms of the relative wavelet energy of
the wave-let coefcients. The energy at each resolution level j is
dened in
0.4
0.6
0.8
1.0
D1 0.4
0.6
0.8
1.0
ED2
(a) (b)
1152 T.H. Loutas et al. / Applied Acoustics 70 (2009) 114811590
25 400 450 500 550 600 650
0.0
0.2
E
# of recordings
(c)
0 25 400 450 500 550 600 650
0.0
0.2
0.4
0.6
0.8
1.0
p12
para
met
er# of recordings
Fig. 6. Parameters evolution during the test for vibr0 25 400
450 500 550 600 650
0.0
0.2
# of recordings
(d)
0 25 400 450 500 550 600 650
0.0
0.2
0.4
0.6
0.8
1.0
p23
para
mete
r# of recordings
ation ch2 (a) ED1, (b) ED2, (c) p12 and (d) p23.
-
(1). The total energy of the wavelet coefcients will then be
givenby:
Etotal Xj
Ej 2
Then; the normalized values are expressed as : pj
Ej=Etotal:3
and the SWE at resolution j is defined as : Hj pj log pj4
Eleven more parameters HD1HD10 and HA1 are then calculated.
4. Test procedure and results
The experimental setup was analytically described in Section 2of
the present work. Many tests were conducted in order to cali-brate
the multi-sensor conguration and assure the repeatabilityof the
recordings and the proper operation with minimum noiseof
acquisition cards, ampliers, pre-ampliers as well as the vari-ous
cables and connections.
Results in terms of various parameters evolution during thetest
from a representative test on a gear system with a transversecut of
25% of root thickness to simulate the tooth crack (Fig. 3) willbe
presented and detailed in this study. Two more tests were con-
ducted on the same conguration yielding similar
parameterbehaviours. Recordings every 5 min were acquired and a
total of650 recordings (54 h of test duration) were resulted until
the ter-mination of the test, that is about 2 h after the tooth was
cut-offthe gear. This type of test was preferred in order to have
the oppor-tunity to monitor both damage modes i.e. the natural gear
wear aswell as the crack propagation, though the latter is dominant
be-tween the two as seen by the minimumwear in the gear faces
afterthe tests. From the recorded vibration and AE waveforms
thewhole set of parametersfeatures as described in Section 3 are
calculated utilizing in-house algorithms developed in
Matlabenvironment. In the following sections the behaviour of the
best from a diagnostic point of view parameters is
analyticallypresented.
4.1. Vibration results
From a total set of about 50 parameters, about 12 of them seemto
have a clear diagnostic potential. For the vibration
recordings,parameters p4, p6, p7, p12, p13, p17, p21, p24, ED1, ED2
and ED3proved capable of attending the damage accumulation upon
thegears and have shown an almost monotonic behaviour duringthe
tests. It is reminded here that p4, p6 and p7 parameters comefrom
the time domain, parameters p12, p13, p17, p21 and p24come from the
frequency domain whereas ED1, ED2 and ED3 arethe wavelet-based
ones. Fig. 4 depicts the evolution of four se-
0.04
0.061st transition
1
0.06
0.08
0.10
oil temeperature effect
2
(a) (b)
T.H. Loutas et al. / Applied Acoustics 70 (2009) 11481159 11530
50 100 150 200 250 300 350 4000.00
0.02
ED
# of recordings
0 50 100 150 200 250 300 350 4000.20
0.25
0.30
0.35
0.40
0.45
0.50
p12
para
met
er
(c) # of recordings
Fig. 7. A magnication of Fig. 4 o0 50 100 150 200 250 300 350
4000.00
0.02
0.04ED
# of recordings
0 50 100 150 200 250 300 350 4000.00
0.01
0.02
0.03
0.04
p23
para
mete
r
(d) # of recordings
ver the rst 420 recordings.
-
lected parameters for ch1 during the test namely ED1, p6, p7
andp13. All parameters shown are normalized in the (01) range.
Atrst sight it seems that no important changes take place
untilapproximately the 420th recording (35 h).
To assist the more accurate observation of the
parametersevolution during this stage of the test, a magnication
was drawnin the diagrams of Fig. 5. In Fig. 5ac a transition in the
regionnear 325th recording (27 h), in the middle of the test, is
ob-served. The size of this rst (of the three transitions
observedthroughout the test) transition depends on the parameter
oneis looking at and in the case of p6 parameter reaches 2200%.
Priorto this point no signicant variation of the parameters are
ob-served. This transition is not evident in Fig. 5d though. A
2ndtransition is identied at the region of the 530th recording(44
h) according to Fig. 4ad. The percentage rise reaches2800% for
parameter p7. A 3rd transition related to the toothcut-off takes
place at about the 625th recording (52 h) asall the graphs in Fig.
4 clearly show. Parameter ED1 sees an in-crease of 700%.
These transitions are important and possess diagnostic value
asthey can be used to dene and characterize critical stages of
thegears damage accumulation and evolution.
Results from the processing of the vibration signals from ch2are
depicted in Fig. 6. The 3rd transition at approximately the
625th recording (52 h) is quite clear whereas the 2nd is not
clearin any of the selected parameters. Looking at Fig. 7 the
location ofthe 1st transition is not very clear as well, at least
not as evident asin the case of vibration ch1, rendering ch2 less
interesting diagnos-tically. Parameter ED1 presented in Fig. 6a
suggests the 1st transi-tion at the neighbourhood of the 325th
recording (27 h). In spiteof the uncertainty and the signicant
uctuations, the aboveparameters are monotonically increased, which
is very useful froma diagnostic point of view.
The area at the very beginning of the test highlighted in Fig.
7bseems to have non-consistent parameter values behaviour, a
phe-nomenon that is attributed to the oil temperature effect upon
therecordings. In the beginning of the test it normally takes few
hoursuntil the lubricant reaches a steady temperature. While the
oiltemperature changes, so does the oil lm thickness between
theasperity contacts of the gears affecting the vibration as well
asthe AE recordings. This is a statement not only valid for the
param-eter ED2 of Fig. 7b. This behaviour in the beginning of the
tests isobserved more or less in almost every parameter presented
inthe paper.
In ch3, a behaviour similar to that of vibration ch1 is
observed.The 2nd and 3rd transitions are clearly dened in the
graphs givenin Fig. 8. The 1st transition is not as clear as in
Fig. 4 graphs but stillcan be marked at least in Fig. 9a and b.
0.2
0.4
0.6
0.8
1.0
(tooth break)
440%
3rd transition
2nd transition
ED1
para
mete
r
0.2
0.4
0.6
0.8
1.0
320%
ED2
para
mete
r
(a) (b)
1154 T.H. Loutas et al. / Applied Acoustics 70 (2009) 114811590
25 400 450 500 550 600 650
0.0600%
# of recordings
0 25 400 450 500 550 600 650
0.0
0.2
0.4
0.6
0.8
1.0
690%
1500%
p7 p
aram
eter
(c) # of recordings
Fig. 8. Parameters evolution during the test for vib0 25 400 450
500 550 600 650
0.0700%
# of recordings
0 25 400 450 500 550 600 650
0.0
0.2
0.4
0.6
0.8
1.0
130%
110%
p12
para
mete
r
(d) # of recordings
ration ch3 (a) ED1, (b) ED2, (c) p7 and (d) p12.
-
cous0.010
0.015
1st transition
eter
(a)
T.H. Loutas et al. / Applied ATable 2 summarizes the percentage
changes of the variousparameters shown in vibration channel 3.
Channel 2 is not in-cluded as the least interesting from the three.
The 1st transitionin some cases is not clear and no value is
given.
4.2. AE results
For the acoustic emission recordings, parameters p4, p6, p7,
p12,p13, p17, p21, p24, ED3, ED4 and ED5 proved capable of
attendingthe damage accumulation upon the gears and have shown an
inter-
0 50 100 150 200 250 300 350 4000.000
0.005ED1
para
m
# of recordings
0 50 100 150 200 250 300 350 4000.0000
0.0002
0.0004
0.0006
0.0008
p7 p
aram
ete
r
# of recordings
(c)
Fig. 9. A magnication of Fig. 6 o
Table 2Parameter percentage changes for vibration channels 1 and
3.
Channel Parameter 1st transition (%) 2nd transition (%) 3rd
transition (%)
1 ED1 25 1500 700p6 2200 1100 350p7 122 2800 450p13 90 300
3 ED1 600 440ED2 700 320p7 1500 690p12 130 1100 50 100 150 200
250 300 350 4000.00
0.02
0.04
0.06
0.08
1st transition
ED2
para
mete
r
# of recordings
0.20
(b)
(d)
tics 70 (2009) 11481159 1155esting monotonic behaviour during
the tests. Parameters p4, p6and p7 are calculated in the time
domain, p12, p13, p17, p21 andp24 in the frequency domain and
parameters ED3, ED4, ED5 arethe wavelet-based ones. AE in total has
a very interesting behav-iour in this test (Fig. 10). Unlike the
results from vibration record-ings in the previous section, even
from the early beginning it has aseemingly linear increasing
behaviour and it seems capable ofdiagnosing even the initial stages
of crack propagation. A closerlook at the rst 400 recordings
reveals a bilinear behaviour moreevident in parameters ED3 and ED5
(see Fig. 11).
It is reminded that the dominant damage mode involved in
thistest is the crack propagation and much less the natural wear.
Thischange in the slope could be associated with changes in the
crackpropagation rate. Parameter ED5 has a diagnostic advantage
sinceits slope changes close to the 150th (12.5 h) recording much
ear-lier than the 250th recording (21 h) of parameter ED3. An
impor-tant transition around the 625th recording (52 h) warns
withrespect to the oncoming tooth failure.
In acoustic emission ch2, interesting diagnostically
behavioursare acquired as Fig. 12 shows. More than two different
slopes canbe identied as Fig. 13 suggests for parameters ED3 and
ED4. Inany case, as in AE ch1, an important transition around the
625threcording (52 h) warns with respect to the oncoming
toothfailure.
0 50 100 150 200 250 300 350 4000.05
0.10
0.15p1
2 pa
ram
ete
r
# of recordings
ver the rst 420 recordings.
-
b)cous1.0(a) (1156 T.H. Loutas et al. / Applied AAE results of
diagnostic parameters coming from ch3 (Fig. 14)seems to have an
almost linear behaviour again, with not signi-cant slope changes
during the test, thus making ch3 behaviourthe least interesting
among the three AE channels. Still the criticaltransition at around
the 625th recording (52 h) is clearly shown.
0 100 200 300 400 500 6000.0
0.2
0.4
0.6
0.8
transition(tooth break)90%
300%
ED3
para
mete
r
# of recordings
ED5
para
mete
r
0 100 200 300 400 500 6000.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
45%
60%
p4 p
aram
ete
r
# of recordings
(d(c)
Fig. 10. Parameters evolution during the test for
0 50 100 150 200 250 300 350 4000.05
0.10
0.15
0.20
0.25
ED3
para
mete
r
# of recordings
(a) (
Fig. 11. A magnication of ED 3 and ED 51.0 tics 70 (2009)
11481159Table 3 summarizes the percentage changes of the various
param-eters presented in all three AE channels. The rst refers to
thechange observed from the test start until the transition and
thesecond refers to the change measured at the neighbourhood ofthe
transition.
0 100 200 300 400 500 6000.0
0.2
0.4
0.6
0.8
1700%
130%
# of recordings
0 100 200 300 400 500 6000.4
0.5
0.6
0.7
0.8
0.9
1.0
30%
50%p12
para
mete
r
# of recordings
)
AE ch1 (a) ED3, (b) ED5, (c) p4 and (d) p12.
0 50 100 150 200 250 300 350 4000.00
0.05
0.10
0.15
0.20
0.25
ED5
para
met
er
# of recordings
b)
over the rst 420 recordings AE ch1.
-
cousT.H. Loutas et al. / Applied AAfter analysing and commenting
on the behaviour of carefullyselected parameters in the previous,
in Fig. 15 an example of the
0 100 200 300 400 500 6000.0
0.2
0.4
0.6
0.8
1.0
35%
100%
ED3
para
mete
r
# of recordings
0 100 200 300 400 500 600
0.0
0.2
0.4
0.6
0.8
1.0
3500%
400%
p6 p
aram
eter
# of recordings
(a) (b
(d(c)
Fig. 12. Parameters evolution during the test for
0 50 100 150 200 250 300 350 400 450 5000.00
0.05
0.10
0.15
0.20
0.25
0.30oil temperature effect
ED3
para
mete
r
# of recordings
(a) (
Fig. 13. A magnication over the rst 500 rectics 70 (2009)
11481159 1157behaviour of non-useful diagnostically- parameters
extracted bythe analysis of AE monitored signals is depicted.
0 100 200 300 400 500 6000.0
0.2
0.4
0.6
0.8
1.0
290%
80%
ED4
para
mete
r
# of recordings
0 100 200 300 400 500 6000.0
0.2
0.4
0.6
0.8
1.0
60%
150%p13
para
met
er
# of recordings
)
)
AE ch2. (a) ED3, (b) ED5, (c) p6 and (d) p13.
0 50 100 150 200 250 300 350 400 450 5000.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
ED4
para
mete
r
# of recordings
b)
ordings for (a) ED3 and (b) ED4 AE ch2.
-
(cous0.8
0.9
1.0ete
r
(a) 1158 T.H. Loutas et al. / Applied A5. Conclusions
The health monitoring of rotating machinery and power
drivetrains is of utmost importance in various industrial
applicationsin industry and in rotorcraft aviation. A single-stage
gearbox wasutilized in order to study the development of damage in
articiallyinduced cracks in the gears. Multi-hour tests were
conducted andnumerous recordings were acquired using acoustic
emission andvibration monitoring. The main goal of the study was to
extract aset of parametersfeatures and check their diagnostic
behaviour
0 100 200 300 400 500 600
0.05
0.10980%
25%
ED5
para
m
# of recordings
0 100 200 300 400 500 6000.2
0.4
0.6
0.8
1.0
55%
51%
p4 p
aram
eter
# of recordings
((c)
Fig. 14. Parameters evolution during the test for
Table 3Parameter percentage changes for AE channels 13.
Channel Parameter Start until transition (%) Transition (%)
1 ED3 300 90ED5 1700 130p4 60 45p12 50 30
2 ED3 100 35ED4 290 80p6 3500 400p13 150 60
3 ED5 25 980ED8 350 870p4 55 51p13 170 1300 100 200 300 400 500
6000.00
0.05
0.10
0.8
0.9
1.0
870%
350%
ED8
para
mete
r
# of recordings
0.8
1.0
b)
d)
tics 70 (2009) 11481159searching for the most potential and
appropriate for future healthmonitoring schemes. A large number of
parameters are proposed.Among them, conventional time domain based
parameters, fre-quency domain based and a set of innovative
parameters basedon the discrete wavelet transform.
Detailed results on the diagnostic behaviour and potentiality
ofthe most interesting of the above parameters/features novel
andconventional were analytically presented and discussed.
Transi-tions in the parameter values were highlighted suggesting
criticalchanges in the operation of the gearbox. Very interesting
behaviourof selected parameters was observed for both
monitoringtechniques.
The oil temperature effect upon vibration and AE recordingswas
clearly identied in the beginning of the tests rendering itan
important factor that should be taken into account in
healthmonitoring of rotating structures. Several features extracted
fromthe recorded vibration and AE waveforms revealed their
variationas the oil temperature was rising up to the
operationaltemperature.
Acoustic emission technique seems superior in the early stagesof
the test and up to the middle being more capable of giving
sig-nicant indications and differentiations to the monitored
parame-ters, something that was not observed for the
vibrationmonitoring. A regionally linear behaviour of AE parameters
wasobserved and the gradients changes were associated with
changesin the crack propagation rate. A superiority of the AE
technique
0 100 200 300 400 500 6000.0
0.2
0.4
0.6130%
170%
p13
para
mete
r
# of recordings
AE ch3. (a) ED5, (b) ED8, (c) p4 and (d) p13.
-
0.5
0.6
0.7
0.8
0.9
1.0p8
par
amete
r
0.2
0.4
0.6
0.8
1.0
ED8
para
mete
r
(a) (b)
tes
T.H. Loutas et al. / Applied Acoustics 70 (2009) 11481159
1159over vibration monitoring regarding the monitoring of early
crackpropagation, is thus concluded.
Acknowledgements
A big part of this research work was conducted in the frame-work
of FP6-European Project ADHER Automated Diagnosis forHelicopter
Engines and Rotating Parts Project.
References
[1] Dempsey PJ. Integrating oil debris and vibration
measurements for intelligentmachine health monitoring. Technical
report, NASA/TM-2003-211307.
[2] Dempsey PJ. A comparison of vibration and oil debris gear
damage detectionmethods applied to pitting damage. Technical
report, NASA/TM-2000-210371.
[3] Dempsey PJ, Mosher M, Huff Edward M. Threshold assessment of
geardiagnostic tools on ight and test rig data. Technical report,
NASA/TM-2003-212220.
[4] Dempsey PJ, Afjeh AA. Integrating oil debris and vibration
gear damagedetection technologies using fuzzy logic. Technical
report, NASA/TM-2002-211126.
[5] Decker Y, Lewicki D. Spiral bevel pinion crack detection in
a helicoptergearbox. Technical report, NASA/TM-2003-212327.
[6] Wang WJ, McFadden PD. Early detection of gear failure by
vibration analysis
0 100 200 300 400 500 6000.4
# of recordings
Fig. 15. Non-interesting parameters evolution during theI.
Calculation of the time-frequency distribution. Mech Syst Signal
Process1993;7(3):193203.
[7] Wang WJ, McFadden PD. Early detection of gear failure by
vibration analysis II. Interpretation of the timefrequency
distribution using image processingtechniques. Mech Syst Signal
Process 1993;7(3):20515.
[8] Wang WJ, McFadden PD. Application of orthogonal wavelets to
early geardamage detection. Mech Syst Signal Process
1995;9(5):497507.
[9] Wang WJ, McFadden PD. Application of wavelets to gearbox
vibration signalsfor fault detection. J Sound Vib
1996;192(5):92739.[10] Baydar N, Ball A. Detection of gear
deterioration under varying load conditionsby using the
instantaneous power spectrum. Mech Syst Signal
Process2000;14(6):90721.
[11] Baydar N, Chen Q, Ball A, Kruger U. Detection of incipient
tooth defect inhelical gears using multivariate statistics. Mech
Syst Signal Process2001;15(2):30321.
[12] Baydar N, Ball A. A comparative study of acoustic and
vibration signals indetection of gear failures using WignerVille
distribution. Mech Syst SignalProcess 2001;15(6):1091107.
[13] Baydar N, Ball A. Detection of gear failures via vibration
and acoustic signalusing wavelet transform. Mech Syst Signal
Process 2003;17(4):787804.
[14] Tandon N, Mata S. Detection of defects in gears by acoustic
emissionmeasurements. J Acoust Emiss 1999;17(1-2):237.
[15] Singh A, Houser DR, Vijayakar S. Early detection of gear
pitting. In: Powertransmission and gearing conference, ASME, DE,
vol. 88; 1996. p. 6738.
[16] Toutountzakis T, Tan CK, Mba D. Application of acoustic
emission to seededgear fault detection. NDT&E Int
2004;37:110.
[17] Tan CK, Mba D. Identication of the acoustic emission source
during acomparative study on diagnosis of a spur gearbox. Tribol
Int 2005;38:46980.
[18] Toutountzakis T, Mba D. Observations of acoustic emission
activity during geardefect diagnosis. NDT&E Int
2003;36:4717.
[19] Tan CK, Irving P, Mba D. A comparative experimental study
on the diagnosticand prognostic capabilities of acoustics emission,
vibration and spectrometricoil analysis for spur gears. Mech Syst
Signal Process 2007;21(1):20833.
[20] Loutas TH, Kalaitzoglou J, Sotiriades G, Kostopoulos V. A
novel approach forcontinuous acoustic emission monitoring on
rotating machinery without theuse of slip-ring. J Vib Acoust
2008;130(6).
0 100 200 300 400 500 6000.0
# of recordings
t for (a) p8 from AE ch1 and (b) ED8 from vibration ch2.[21]
Kostopoulos V, Loutas T, Derdas C, Douzinas E. Wavelet analysis of
headacceleration response under Dirac excitation for early oedema
detection. TransASME J Biomech Eng 2008;130(2).
[22] Loutas TH, Kostopoulos V. Health monitoring of
carbon/carbon wovenreinforced composites. Damage assessment by
using advanced signalprocessing techniques. Part II:
acousto-ultrasonics monitoring. Compos SciTechnol
2009;69(2):27383.
Condition monitoring of a single-stage gearbox with artificially
induced gear cracks utilizing on-line vibration and acoustic
emission measurementsIntroductionExperimental setupSignal
processing methodologiesConventional parametersDiscrete wavelet
transform based parameters
Test procedure and resultsVibration resultsAE results
ConclusionsAcknowledgementsReferences