Sie sind auf Seite 1von 10

GEAR FAULT MONITORING:

COMPARISON OF VIBRATION ANALYSIS TECHNIQUES

G. DALPIAZ, A. RIVOLA and R. RUBINI


Dipartimento di Ingegneria delle Costruzioni Meccaniche, Nucleari, Aeronautiche e di Metallurgia - University of Bologna
Viale Risorgimento, 2 - I-40136 Bologna - Italy
giorgio.dalpiaz@mail.ing.unibo.it

ABSTRACT

This paper deals with gear condition monitoring based on vibration analysis techniques. The detection and
diagnostic capability of some of the most effective techniques are discussed and compared on the basis of
experimental results, concerning a gear pair affected by a fatigue crack. In particular, the results of new
approaches based on time-frequency and cyclostationarity analysis are compared against those obtained by
means of the well accepted cepstrum analysis and amplitude and phase demodulation of meshing harmonics.
Moreover, the sensitivity to fault severity is assessed by considering two different depths of the crack. The effect
of choosing different transducer locations and different processing options are also shown.
In the case of the experimental results considered in this paper, the power cepstrum is practically insensitive to
the crack evolution. Conversely, the Spectral Correlation Density function is able to monitor the fault
development and does not seem to be significantly influenced by the transducer position. The demodulation
techniques are able to localise the damaged tooth; however, their sensitivity is strongly dependent on the proper
choice of the filtering band and is affected by the transducer location. The Wavelet transform seems to be a good
tool for crack detection; it is particularly effective if the residual part of the time synchronous averaged signal is
processed.

1 - INTRODUCTION

Most modern techniques for gear diagnostics are based on the analysis of vibration signals picked up from the
gearbox casing. The common target is to detect the presence and the type of fault at an early stage of
development and to monitor its evolution, in order to estimate the machines residual life and choose an adequate
plan of maintenance. It is well known that the most important components in gear vibration spectra are the tooth
meshing frequency and its harmonics, together with sidebands due to modulation phenomena. The increment in
the number and amplitude of such sidebands may indicate a fault condition. Moreover, the spacing of the
sidebands is related to their source [1].
The simple spectral analysis is generally unable to detect gear failures at an early stage; for this reason, many
researchers have proposed the application of other vibration analysis techniques for the early detection of fault
symptoms. The aim of this paper is to assess and compare the detection and diagnostic capability of some of the
most effective techniques, on the basis of experimental results.
Cepstral analysis has been widely applied to gear monitoring. The cepstrum is well suited for the detection of
sidebands in vibration spectra and for the estimation of their evolution during gear life. In addition, since the
cepstrum estimates the average sideband spacing over a wide frequency range, it allows very accurate
measurement of the sideband periodicity. It is therefore applicable to both detection and diagnosis of gear faults
[1-3].
The amplitude and phase demodulation of one of the tooth meshing harmonics is a well known gear monitoring
technique [4, 5]. This technique requires the time synchronous averaging of the vibration signal in order to
remove any periodic events not exactly synchronous with the gear of interest and to reduce the effects of noise
and vibration sources other than gear pairs [1, 4]. The averaged signal is then bandpass filtered around one of the
larger meshing harmonics and the amplitude and phase modulation are obtained by means of a procedure based
on the Hilbert transform.
The cyclostationary process theory has been recently applied to gear monitoring. In particular, the Spectral
Correlation Density (SCD) function of gear vibration indicates the correlations between meshing harmonics

623
and their sidebands [6, 7], that are just the spectral characteristics primarily affected by gear faults. Thus the SCD
function permits gear fault detection and identification of the damaged element.
Local faults in gears (e.g. crack in a gear tooth) produce impacts [1]. As a result of this excitation, transient
modifications of vibration signals may be observed. Therefore, the vibration signal can be considered as non-stationary.
However, most of the widely-used signal processing techniques, are based on the assumption of stationarity and globally
characterise signals. Thus they are not fully suitable for detecting short-duration dynamic phenomena; as a matter of
fact, the time-localisation of transient events is impossible. On the other hand, the application of time-frequency
distribution techniques - such as Wavelet Transform (WT) - is highly suitable [8-10]. By means of this time-variant
method it is possible to detect and localise the presence of cracks in gears.
In this paper, the above-mentioned analysis techniques are applied to experimental vibration data, concerning a gear pair
affected by a fatigue crack at the root of one of the teeth. The capability of fault detection and diagnosis are discussed
and compared; in particular, the sensitivity to fault severity is assessed by considering two different depths of the crack.
The effect of choosing different transducer locations and different processing options are also shown.

2 - TESTS

Tests were performed on a power circulating gear testing machine composed of two identical single-stage gear units
mounted back to back, with a locked-in torque [11]. Each gear unit contains a carburised spur gear pair of module 3
mm; the pinion and the wheel have respectively 28 and 55 teeth. Further data about the gears can be found in [5]. Tests
were carried out after the introduction of a crack at the root of one tooth of the wheel mounted in one of the gear units.
The tooth was precracked before mounting the gear on the testing machine, by using an appropriate device in order to
apply a fatigue load to the tooth in a similar way to when the tooth is loaded during meshing. The crack affected the
whole tooth flank, extending between the two wheel faces.
In order to investigate the influence on the vibration signal of both the transducer location and the development stage of
the crack, the signals were picked up on a gearbox casing in two different positions and two crack dimensions were
examined. In particular, two vibration signals were simultaneously measured from the casing of the damaged gearbox
unit by means of two Brel & Kjr 4369 piezoelectric accelerometers: one was mounted with the sensitivity axis
parallel to the shaft axis ("axial" accelerometer); the other one was placed on the wheel bearing casing in a radial
direction ("radial" accelerometer) [5]. The signals were integrated in Brel & Kjr 2635 charge preamplifiers to obtain
velocity signals. In addition, a one-per-wheel revolution tachometer signal was taken using an inductive proximity
probe. Tests were subsequently performed with two crack lengths, corresponding to about 20% and 45% of whole
fracture surface after breakage.
The results presented in this paper are relative to a nominal pinion torque of 385 Nm and nominal pinion speed of 1000
rpm (16.67 Hz); thus the meshing frequency is 466.67 Hz. It is worth noting that the actual operating conditions were
slightly different in the tests carried out with the two crack dimensions.
The signals were recorded on magnetic tape and converted to a digital time-series using a DIFA Scadas unit having a
12-bit ADC with variable sampling frequency, which was driven by the LMS Cada-X software. The analysis was limited
to the frequency range 0-5000 Hz which includes the most meaningful meshing harmonics. The digital signals were
processed and analysed by means of the LMS Cada-X and MATLAB softwares. Whenever it was required, the signals
were pre-processed by means of a synchronising technique in order to compensate slight - but quite important -
variations of the machine angular speed [12]. For this purpose, a particular synchronising methodology was used [13,
14]. The synchronised signal records have 1024 points per wheel revolution. Some techniques were also applied to the
signals, after a process of time synchronous averaging over 28 revolutions of the damaged gear (i.e. the wheel); as
mentioned, this averaging reduces the effects of vibration sources other than the wheel.

3 - EXPERIMENTAL RESULTS AND DISCUSSION

3.1 - Cepstrum analysis

As mentioned in the introduction, the gear vibration spectra commonly show families of sidebands of the meshing
harmonics. For gearboxes in good condition the sideband level generally remains constant with time. Changes in the
number and strength of the sidebands normally indicate a deterioration condition. Such sidebands typically arise from
the modulation of the gear vibration by the rotational frequency, as in the case of a crack in one of the teeth (there is a
fault once per revolution). Therefore, a family of sidebands has the same spacing, that is the fundamental modulating
frequency. Consequently, the sideband spacing contains diagnostic information, since it is related to the modulation
source [2]. However, it can be difficult to distinguish and evaluate the sideband spacing by means of spectral analysis,
due to its poor resolution. In order to overcome this problem, cepstral analysis can be employed. Various forms of
cepstrum exist, but all can be considered as a spectrum of a logarithmic (amplitude) spectrum [2]. The cepstrum is

624
therefore useful in interpreting the spectrum structure and in detecting the spectrum periodicity [1-3]. The quefrency of
the cepstrum peaks represents the modulation period and its reciprocal the modulating frequency.
The power cepstrum of the experimental signals was computed from a 2048-line power spectrum in the range 0-4344
Hz, obtained with 25 spectral averages. The results reported in Fig. 1 (where the amplitude of the cepstrum is plotted),
are relative to the case of the radial vibrations. Peaks at about 0.06 s (16.67 Hz) and about 0.12 s, are present; they
correspond to the first rahmonic of the pinion and the wheel, respectively. Although the signals are relative to different
crack dimensions, their cepstra are very similar (see Fig. 1). Moreover, the cepstral component corresponding to the first
rahmonic of the damaged gear (0.12 s), is practically insensitive to the crack dimension. In other words, for this case,
the cepstrum analysis gives unclear monitoring information and is not able to detect the presence of the damage. The
cepstrum approach was also employed for the axial vibration signals, but the results were not comforting and have not
been reported for the sake of brevity.
It was suspected that variations of the machine angular speed were the cause of such results. As a matter of fact, such
variations could alter the sidebands during the FFT computation. Therefore, the signals were pre-processed by means of
a synchronising technique and the cepstral analysis was repeated, employing a 2048-line power spectrum in the range 0-
4344 Hz (7 averages were used).

3 3
(a) (b)
Amplitude [dB]

Amplitude [dB]
2 2

1 1

0 0.06 0.12 0.18 0.24 0 0.06 0.12 0.18 0.24


Quefrency [s] Quefrency [s]

Fig. 1 - Cepstrum of the radial vibration: small (a) and large (b) crack.

8 8
(a) (b)
Amplitude [dB]

Amplitude [dB]

6 6

4 4

2 2

0 0
0 180 360 540 720 0 180 360 540 720
Quefrency (Wheel Rotation [deg]) Quefrency (Wheel Rotation [deg])

Fig. 2 - Cepstrum of the radial vibration (synchronised signal): small (a) and large (b) crack.

Fig. 2 shows the results for the radial vibration. In comparison to the non-synchronised signals, the first rahmonic of the
wheel is enhanced, whilst that of the pinion almost disappears. Although the first rahmonic of the damaged gear now
dominates the cepstrum, the fault monitoring cannot still be performed, since the cepstrum is insensitive to the crack
increment.
The cause of the cepstrum inefficiency could be due to the fact that speed and torque were slightly different in the tests
carried out with the two crack dimensions. As a matter of fact, it was found that the different operating conditions
altered the vibration spectra; further investigation is thus needed.

3.2 - Demodulation analysis

The effects of local faults, such as cracked teeth, on the amplitude and phase modulation of the meshing harmonics are
widely discussed in [1]. In [4], the author proposed a signal processing methodology based on the demodulation of the
amplitude and phase of one of the meshing harmonics of the signal, as a gear monitoring technique. That method was
further developed in [5] in order to increase its sensitivity. The time synchronous average (TSA) has to be performed
before applying the demodulation approach. As a matter of fact, the TSA process enhances the signal components due to
the gear of interest and strongly reduces the effects of all other sources, and the noise. In addition, as the TSA is exactly

625
periodic, several operations such as filtering and signal component elimination can be easily carried out in the frequency
domain. As a drawback, it is necessary to repeat the analysis for each gear, for a complete machine diagnosis.
The demodulation was applied to the experimental vibration signals by filtering about the third meshing harmonic (at
165 wheel shaft orders), since this was the most important and showed the strongest modulation sidebands. In particular,
30 upper sidebands of the third meshing were extracted for the analysis.

1.5 1.5

1 1

0.5 0.5
(a) (b)
0 0
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

1.5 1.5

1 1

0.5 0.5
(c) (d)
0 0
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

Fig. 3 - Amplitude modulation (165-195 orders); axial vibration: small (a) and large (b) crack;
radial vibration: small (c) and large (d) crack.

10 10
0 0
-10 -10
-20 -20
-30 -30
(a) (b)
-40 -40
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

5 5

0 0

-5 -5

-10 -10
(c) (d)
-15 -15
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

Fig. 4 - Phase modulation derivative (165-195 orders); axial vibration: small (a) and large (b) crack;
radial vibration: small (c) and large (d) crack.

626
Fig. 3 presents the results of the amplitude demodulation process applied on both the axial and radial velocity signals
after the time synchronous averaging over 28 revolutions of the wheel. Both the crack dimensions are examined. The
amplitude modulation is able to localise the damaged tooth and shows a good sensitivity to the crack severity; however,
in the case of the radial signal, it is not possible to the detect the small crack.
The crack also affects the phase modulation which is characterised by a lag due to the deflection of the affected tooth as
it meshes with the other gear. In particular, as established in other tests performed on the same gear system, the earlier
phase modulation symptom is its sudden variation; thus, the phase modulation derivative (PMD) was introduced to make
crack detection easier [5].
For the same signals as in Fig. 3, the PMD is reported in Fig. 4. The peaks in the PMD clearly reveal the presence and
position of the crack. The PMD approach is more sensitive than the amplitude one; however, the small crack in the case
of radial vibration is hardly detected by this tool either. Therefore one can say that the transducer location affects the
results of the demodulation analysis.
In order to give an indication about the damage severity, a statistical parameter of the PMD can be employed; here the
kurtosis is used. For the axial signal, the ratio of the kurtosis value in the case of large crack to the value for the small
crack is 5.7; this ratio is 5.8 in the case of the radial vibration.
The results of the demodulation approach are strongly dependent on the filtering band, as discussed in [5]. As an
example, Fig. 5 shows the phase modulation derivative of the radial signal obtained by bandpass filtering the TSA in the
range 55 to 85 wheel shaft orders (right sidebands of the first meshing harmonic). Thus, this technique gives good
results on the condition that a proper frequency band is chosen.

5 5

0 0

-5 -5

-10 -10
(a) (b)
-15 -15
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

Fig. 5 - Phase modulation derivative (55-85 orders); radial vibration: small (a) and large (b) crack.

3.3 - Cyclostationary analysis

Since the spectral structure of a gear vibration signal is mainly characterised by the interaction between the meshing
harmonics and their sidebands, the cyclostationary process theory has been recently applied to gear monitoring [6,7]. As
a matter of fact, by means of the cyclostationary approach it is possible to investigate the correlation degree between
different frequency components of the spectrum or, in other words, to establish whether they are related each other or
not [15].
In particular, the SCD function of a signal (which can be defined as the Fourier transform of the Fourier coefficient of
the signal autocorrelation function) can be employed to analyse the correlation between a meshing harmonic and one of
its modulating sidebands. Such a sideband is associated with a specific gear element, being a multiple of its revolution
frequency. Therefore, the analysis of the gear spectrum evolution by means of the SCD function allows both the
detection and diagnosis of gear fault.
The SCD function of the experimental signals was computed and the relationship between the third meshing harmonic
and the pertinent first upper modulating sideband, relative to both the gear elements, were investigated. The
synchronised signal records were used; they included 28 wheel revolutions in order to ensure an adequate cyclic
frequency resolution (0.303 Hz). The FFT length employed during the SCD computation was 512 points (i.e. resolution
frequency is 16.97 Hz). The results are shown in Fig. 6, which plots the SCD magnitude over the bifrequency plane (f,
). The correlation between the third meshing harmonic (fm=1400 Hz) and the revolution frequency of the pinion (P
=16.67 Hz) is located at the bifrequency (fm +P/2, P); the SCD value at the point (fm +W/2, W) is a measure of the
correlation between the meshing harmonic and the wheel revolution frequency (W =8.48 Hz). Further details about the
interpretation of the SCD plot can be found in [7].

627
It is noteworthy that peaks are present also for the gear in sound conditions (the pinion); thus, the SCD function can only
be employed to monitor the fault evolution. However, by observing the amplitude modification of the correlation peaks,
one can conclude that the damage affects the wheel. As a matter of fact, the peak that corresponds to the wheel shows a
strong increment with the crack severity, whilst the pinion correlation peak only exhibits slight changes, that might be
due to the mentioned small differences in the operating conditions. In addition, the transducer position does not seem to
significantly influence the results.

(a) (b)

(c) (d)

Fig. 6 - Spectral Correlation Density between the third meshing harmonic and its upper modulating
sideband; axial vibration: small (a) and large (b) crack; radial vibration: small (c) and large (d)
crack.

3.4 - Wavelet transform

The presence of a crack in one tooth introduces short-duration changes in the vibration signal. On the contrary, more
distributed faults (e.g. geometrical imperfections in the gear train and uniform wear), introduce "slow" modifications of
the signals over the revolution period. For advanced local faults, time domain techniques may be sufficient to detect the
damage, but the early detection of defects requires more sophisticated signal process methods. The non-stationary nature
of the signal suggests the use of time-frequency techniques, which make it possible to look at the time evolution of the
signals frequency content. Many such time-variant methodologies exist; in [8-10], the authors successfully applied the
WT technique to the detection of cracked teeth in gears.
In order to examine the sensitivity of the method, the WT was applied to the radial vibration in the case of the small
crack dimension; as a matter of fact, this signal seems to be scarcely affected by the fault, as has been established by
means of the demodulation technique [see Fig. 3(c) and 4(c)].
Fig. 7 reports the WT amplitude map corresponding to one wheel revolution; the analysed frequency range is 500-3000
Hz, which includes the most important meshing harmonics. The transient event should cause an increment of a WT
amplitude of short duration over a wide frequency range. Unfortunately, that does not happen, making the result useless.
The signal was probably affected by noise or other vibration sources; its TSA was therefore subsequently considered.
The result does in fact improve, as is shown in Fig. 8(a), where a localised increment of the WT amplitude is visible at

628
about 150 degrees of the wheel rotation. The presence and the angular position of the crack can be detected only by
carefully inspecting the WT cross-sections [see Fig. 8(c)]; however, the crack effects are evident only in some frequency
ranges [10], thus a fault detection procedure would be strongly dependent on the choice of the cross-section frequency.
In order to overcome this limit, the residual approach [16-18] was applied to the radial TSA signal before computing the
WT. The residual signal can be obtained by removing the harmonic part of the original signal, and it represents the
amount of the original signal departure from its ideal counterpart.

Fig. 7 - Wavelet Transform map of the raw radial vibration for the small crack dimension (one wheel revolution).

0.05 2
(c) (d)
WT cross-section at 2225 Hz sum of the WT cross-sections
Amplitude

Amplitude

0.025 1

0 0
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

Fig. 8 - WT of the TSA radial vibration for the small crack dimension: original signal (left);
residual signal (right). (a) and (b) WT map; (c) WT cross-section at 2225 Hz; (d) sum of the WT
cross-sections (500-3500 Hz).

629
Fig. 8(b) shows that the residual signal is a more sensitive diagnostic tool; in fact, by observing its WT map, it is
possible to clearly distinguish the transient effects introduced by the cracked tooth; conversely, in the original signal
they were obscured by the harmonic components. Moreover, such a procedure makes it possible to localise the damage
whatever the frequency section of the WT map. In particular, the sum of the WT cross-sections [Fig. 8(d)] can be used
in order to make a feature extraction procedure easy.
Fig. 9 shows the results obtained by computing the WT of the residual of TSA signals: both the transducer locations are
examined (axial and radial vibration) for the two crack dimensions. As a measure of the crack severity, the fourth
statistical central moment was evaluated for the four cases of Fig. 9: in the case of the axial vibration, the ratio of this
statistical indicator value in the case of the large crack to the value for the small crack is 2.7; the ratio is 26.7 in the case
of the radial vibration.

5 5
(a) (b)
4 4
Amplitude

Amplitude
3 3
2 2
1 1
0 0
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

3 3
(c) (d)
Amplitude

Amplitude

2 2

1 1

0 0
0 360 0 360
Wheel Rotation [deg] Wheel Rotation [deg]

Fig. 9 - Sum of the WT cross-sections (500-3500 Hz) of the residual of TSA signal; axial vibration:
small (a) and large (b) crack; radial vibration: small (c) and large (d) crack.

4 - CONCLUSIONS

This paper compares the effectiveness of some vibration analysis techniques for detection and diagnostics of cracks in
gear teeth, on the basis of experimental results. In particular, the capability of new approaches based on time-frequency
and cyclostationarity analysis are compared against those obtained by means of the well accepted cepstrum analysis and
amplitude and phase demodulation of meshing harmonics; in addition, different processing options are considered in
some cases. For each technique, the sensitivity to crack depth and the influence of the transducer position are assessed.
Applying cepstrum analysis and Spectrum Correlation Density, it is possible to monitor each gear element in a gearbox
at the same time, by observing amplitude modifications of the correspondent quefrency and correlation peak. On the
other hand, these techniques give only information about spectrum evolution; in other words, they are unable to detect
faults from the analysis of only one vibration measurement, as peaks are present also for gears in sound conditions; thus
the evolution during machine operation must be considered.
In the case of the experimental results considered in this paper, the power cepstrum of both the raw vibration signals and
the synchronised signals was considered; in both cases this technique is practically insensitive to the crack dimension,
i.e. to the fault evolution. On the other hand, the SCD of the synchronised signals is very sensitive to the fault evolution,
as the peak corresponding to the wheel shows a strong increment with the crack dimension, whilst the pinion correlation
peak only exhibits slight changes. In addition, the SCD does not seem to be significantly influenced by the transducer
position.

630
For the evaluation of the amplitude and phase modulations of one of the tooth meshing harmonics, the time synchronous
average - relative to the monitored gear - must be previously computed; thus, it is necessary to repeat the analysis for
each gear, in order to obtain a complete machine diagnosis. According to the presented results, both amplitude and
phase modulation are sensitive to the presence and dimension of the crack. In particular, the phase modulation derivative
exhibits high sensitivity, as a sudden phase variation is the earlier modulation symptom. Moreover, this technique is able
to localise the damaged tooth. However some drawbacks were found: the effectiveness and the sensitivity of this
technique strongly depend on the proper choice of the processed frequency band and, in addition, the sensitivity is
influenced by the transducer position. For these reasons, the practical application of this technique may be sometimes
ineffective.
As faults localised in one or a few teeth produce transient dynamic effects, the application of the Wavelet Transform
appears well suited. This technique makes it possible to localise the damaged tooth. The presented results show that the
WT of the raw signals is practically insensitive to cracks, while the sensitivity of the WT of TSA signals is quite
satisfactory. Thus, for a complete diagnosis, it is necessary to repeat the analysis with reference to each gear in the
machine. Moreover, as the crack effects are evident only in some frequency range, a fault detection procedure would be
strongly dependent on the choice of a proper cross-section in the WT map. This limit is overcome if the residual of the
TSA signal is processed by the WT. The expounded approach proves itself indispensable in monitoring small crack
effects - which are often covered by harmonic components of the signal - and makes it possible to localise the damage
over a wide range of frequency in the WT map. Consequently, the sum of the WT cross-section is a useful tool for crack
detection: as a matter of fact, it is very sensitive to both the presence and the dimension of the crack, even if its
sensitivity fairly changes for different transducer positions.
A successful diagnostic tool should undoubtedly integrate several of the currently available techniques; in this context,
the SCD approach and the WT of the residual part of the TSA signal appear to be very suitable techniques for crack
detection in gear.

ACKNOWLEDGEMENTS
This work was partially supported by a grant from the CNR - Italian National Research Council.

REFERENCES

1. RANDALL R.B. 1982 Journal of Mechanical Design 104, 259-267. A new method of modelling gear faults.
2. RANDALL R.B., HEE J. 1981 Brel & Kjr Technical Review, No. 3, 3-40. Cepstrum analysis.
3. SIDAHMED M. 1990 Bulletin S.F.M. Revue Franaise de Mcanique 4, 243-254. Dtection prcoce de dfauts dans
les engrenages par analyse vibratoire.
4. MCFADDEN P.D. 1986 ASME J.of Vibration, Acoustics, Stress, and Reliability in Design 108, 165-170. Detecting
fatigue cracks in gear by amplitude and phase demodulation of the meshing vibration.
5. DALPIAZ G. 1990 sterreichische Ingenieur - und Architekten - Zeitschrift (IAZ) 135, 312-317. Early detection
of fatigue cracks in gears by vibration analysis techniques.
6. CAPDESSUS C., SIDAHMED M., LACOUME J.L. 1995 Proceedings of the 2nd International Symposium on
Acoustical and Vibratory Surveillance Methods and Diagnostic Techniques, Senlis, France, 391-401. Apport de la
theorie des processus cyclostationnaires a l'analyse et au diagnostic des engrenages.
7. RUBINI R., SIDAHMED M. 1997 Proceedings of the Symposium on Fault Detection, Supervision, and Safety for
Technical Processes, 977-982. Hull, UK. Diagnostics of gear systems using the Spectral Correlation Density of the
vibration signal.
8. STASZEWSKI W.J., TOMLINSON G.R. 1994 Mechanical Systems and Signal Processing 8(3), 289-307. Application
of the Wavelet transform to fault detection in a spur gear.
9. MCFADDEN P.D. 1994 Proceedings of an Internation Conference on Condition Monitoring, Swansea, UK, 172-
183. Application of the wavelet transform to early detection of gear failure by vibration analysis.
10. DALPIAZ G., RIVOLA A., RUBINI R. 1996 Proceedings of the Congress of Technical Diagnostics, Gdansk, Poland
2, 185-192. Dynamic modelling of gear system for condition monitoring and diagnostics.
11. DALPIAZ G., U. MENEGHETTI 1991 Proceedings of 3rd International Conference on Condition Monitoring,
Windsor, UK. Ed. McEwan J.R., London & N.Y.: Elsevier Applied Science, 73-82. Detection and modelling of
fatigue cracks in gears.
12. MCFADDEN P.D. 1989 Mechanical Systems and Signal Processing 3(1), 87-97. Interpolation technique for time
domain averaging of gear vibration.
13. RUBINI R. 1998 Pub. DIEM, University of Bologna, No. 95. Tecnica di sincronizzazione di segnali periodici per il
calcolo della media temporale sincrona in presenza di fluttuazioni della frequenza fondamentale - Parte prima:
Presentazione del metodo (in italian).

631
14. RUBINI R. 1998 Pub. DIEM, University of Bologna, No. 96. Tecnica di sincronizzazione di segnali periodici per il
calcolo della media temporale sincrona in presenza di fluttuazioni della frequenza fondamentale - Parte seconda:
Applicazione agli ingranaggi (in italian).
15. GARDNER W.A. 1986 Introduction to Random Processes with Applications to Signals and Systems. New York:
Macmillan.
16. STEWART R.M. 1977 Institute of Sound and Vibration Research, Paper MHM /R /10 /77. Some useful data
analysis techniques for gearbox diagnostics.
17. BOULAHBAL D., M.F. GOLNARAGHI AND F. ISMAIL 1997 American Gear Manufacturers Association - Technical
paper 97FTM3. Detection of fatigue cracks in gears with the continuous wavelet transform.
18. MCFADDEN P.D. 1987 Mechanical Systems and Signal Processing 1(2), 173-183. Examination of a technique of
failure in gears by signal processing of the time domain average of meshing vibration.

632

Das könnte Ihnen auch gefallen