Mechanical Systems and Signal Processing 42 (2014) 368–376

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Gear parameter identification in a wind turbine gearbox using

vibration signals
Nader Sawalhi a,n, Robert B. Randall b
a
Mechanical Engineering Department, Prince Mohammad Bin Fahd University (PMU), Al Khobar 31952, Saudi Arabia
b
School of Mechanical and Manufacturing Engineering, University of New South Wales, Sydney, NSW 2052, Australia

a r t i c l e in f o abstract

Article history: When carrying out vibration-based diagnosis of gearboxes it is desirable to know the
Received 29 July 2013 numbers of teeth on all gears, so as to be able to calculate toothmesh frequencies and
Received in revised form rotational speeds of all shafts. If the speed varies, this information must be obtained in the
13 August 2013
form of “shaft orders” related to the input and/or output speed. This paper describes how
Accepted 17 August 2013
Available online 1 October 2013
it was possible to extract most of this information from the vibration signal itself in the
case of a wind turbine gearbox with one planetary and two helical parallel stages. Using a
Keywords: spectrogram, a section of signal was first found with minimal speed variation (about 4%)
Harmonics after which the instantaneous speed information was extracted by frequency demodula-
Sidebands
tion of dominant speed related components. After order tracking based on this it was
Gear mesh frequency
found possible to determine the numbers of teeth in the two parallel stages, using very
Order tracking
Spectrogram accurate harmonic cursors applied to each of the shafts of pairs of gears in mesh (with
Wind turbines common mesh frequency). This was successful for the two parallel stages, but the
proposed estimates of the tooth numbers in the planetary section are subject to some
doubt. Allowable combinations are quite restricted using the normally applied rules, but
there can be exceptions. Even so, the presented approach is confirmed as a viable method.
& 2013 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
2. Experimental data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
3. Data processing stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
7. Appendix A. Supplementary material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
8. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

n
Corresponding author. Tel.: þ966 552186218; fax: þ 966 3 8964566.
E-mail addresses: nsawalhi@pmu.edu.sa, nadersaw@hotmail.com, nadersaw@gmail.com (N. Sawalhi), b.randall@unsw.edu.au (R.B. Randall).

0888-3270/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ymssp.2013.08.017
 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376 369

1. Introduction

Monitoring the health of gears in a gearbox, using vibration signals, requires a knowledge of the input or the output
rotational speed, the number of stages as well as their arrangement and the number of teeth for each gear. Often, this
information is available to the vibration analyst by means of a speed reference signal and the manufacturer's details of the
gearbox. Typically, a tachometer/encoder is installed at the high speed shaft to enable estimation of the shaft speed and
order tracking of the signal [1] to remove speed fluctuations.
In cases where more limited information is available to the analyst, a harmonic cursor approach has been recommended for
the blind determination of the numbers of teeth on a gear pair [2]. The use of a harmonic cursor in such occasions is possible if
the gear pair represents a ‘hunting tooth’ design [2]. The ‘hunting tooth design’, which is considered as good practice, means that
there is no common factor between the pinion and the gear and consequently their harmonics are completely separable except at
the gear mesh frequencies (the closest the other harmonics can reach is 1/(m  n), where m and n are the numbers of teeth on
the two gears. In order to use a harmonic cursor, the machine speed is required to be stable to about 1:20000 [2]. If this is not the
case, order tracking can be used to increase the degree of stability. The harmonic cursor approach also requires the presence of a
reasonable number of harmonics of both shafts in the spectrum (logarithmic/decibel scale) including sidebands around the gear
mesh frequency. The harmonic cursor approach as described in [2] starts by setting up a harmonic cursor on each shaft in
succession, first on the low orders and then progressively adjusting it by zooming in higher frequency bands. This will eventually
determine the fundamental frequency to the required accuracy. If lists of the two harmonic series are then compared, the gear
mesh frequency corresponds to where they match to better than 1:10000 [2]. Even if the design is not hunting tooth, the
approach can often still be used. The most likely common factor is 2 or 3, in which case the first correspondence will be at 1/2 or
1/3 of the actual mesh frequency. For the pinion, this will often lead to an unlikely minimum number of teeth, and in any case if
there is an inspection port, the tooth pitch can be measured sufficiently accurately with a tape measure to exclude incorrect
possibilities.
The situation presented in this paper discusses the case where knowledge about the characteristics of the gearbox, of a
wind turbine, was very limited and there was no speed signal. The challenge imposed in such a situation requires an
accurate estimate of the different shaft speeds (or at least their ratios) and the numbers of teeth for each of the gears to
calculate the gear mesh frequencies. The case is complicated by the fact that the wind turbine does not run at constant
speed and in fact the measurements analyzed are actual signals from a full size wind turbine running under normal
conditions and being subject to varying wind conditions and with no tachometer/speed reference signal available. In order
to tackle these issues a signal processing approach has been introduced. The aim of this processing is to arrive at an order
tracked signal suitable for the harmonic/sideband cursor tuning. The processing approach, which is detailed in Section 3
starts by down sampling (decimating) the original signal so that analysis is made in the low frequency region up to 2500 Hz
(dominated by gearmesh frequencies) instead of the full 20 kHz range. The decimated signal was then examined using the
spectrogram to observe speed variation and decide on a typical section with small speed fluctuations for further processing.
Order tracking was then used on the selected section to remove speed fluctuations. This has to be carried out by extracting a
pseudo tachometer from the signal itself. Finally a very accurate harmonic/sideband cursor was employed in an attempt to
predict the number of teeth on each gear.
The paper is organized as follows: after this introduction, the experimental data collected is described in Section 2. This is
followed by an illustration of the data handling and processing in Section 3. Results are presented and discussed in Section 4.
Finally the paper is concluded in Section 5.

2. Experimental data

The case presented in this paper was taken from an actual wind turbine gearbox. The gearbox has three stages (two
helical parallel stages and one planetary). A typical schematic presentation to show the layout of the gearbox is presented in

Fig. 1. Typical Layout of a wind turbine gearbox [3] for illustration purpose only.
370 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376

Fig. 1 [3]. The only information available about this gearbox was that it had three stages and an overall gear ratio close to
78.472. There was no speed reference signal. The generator was of the doubly fed induction type, for which the speed can
vary by up to 730%.
Ten minutes of vibration data, sampled at 51,200 Hz were measured by the company Mechanical Solutions Inc. (MSI)
from a number of accelerometers mounted on the wind turbine. Four accelerometers were placed as follows: one on the
gearbox first stage (planetary), a second on the gearbox third stage (close to the high speed shaft) and the other two on the
inboard and outboard bearings of the generator. No other information or data were made available. A detailed analysis of
one of the signals measured on the gearbox was made in order to attempt to discover the numbers of teeth on the various
gears as is explained in Sections 3 and 4. The signal measured at the Gearbox 3rd Stage (High speed shaft) was chosen
for this.

3. Data processing stages

The different stages of processing the signal are described in Fig. 2. In the first stage, the signal from the third gearbox
stage was decimated (down-sampled) by a factor of 8, partly to allow analysis of longer records (in terms of time) and partly
because the information about the various gear mesh frequencies was contained in the valid frequency range to 2500 Hz
after down-sampling. Matlabs function “decimate” was used for this purpose. Decimation reduces the original sampling rate
to a new lower rate by the use of a lowpass filter and then by resampling the resulting smoothed signal at the lower rate.
The decimation process is opposite to interpolation.
The second stage utilized a Short Time Fourier Transform (Spectrogram). The main purpose of utilizing a spectrogram at
this stage was to observe the speed variation within the measured record and use this to decide on a part where speed
variation is small. The Spectrogram also gives very useful information about the different sets of gear mesh frequencies and
shaft speeds. Matlabs function “Spectrogram” was used at this stage with a hamming window of 8192 samples and 50%
overlap giving a frequency resolution of 0.7813 Hz/line.
Order tracking [1] was used in the third stage to remove speed fluctuations from the signal and enable the tuning of
harmonic cursors to a very high accuracy for a confident analysis. As no tachometer signal was provided, the signal itself was
used to perform the order tracking. In cases where a tachometer signal is not available, a reference signal can be obtained
from the vibration signal itself and used for order tracking [4–7] The approach used in this paper was developed by [7] and
is based on phase demodulation, with the possibility to improve accuracy by successive iteration, perhaps starting with a
very low harmonic for which a much larger speed variation can be handled. In the current application, phase demodulation
around the first harmonic of the output shaft speed was initially performed and was then refined by demodulating around
the high speed stage gear mesh frequency. After the order tracking process, which removes the speed fluctuations from the
signal, harmonics of different shaft families and gear mesh families become recognizable, and extremely localised.
A harmonic cursor [2] could then be used to identify the different harmonic and sideband families, and thus determine
the numbers of teeth on each gear. Note that in a parallel shaft gearbox, modulation sidebands are also harmonics and so
can be identified using a harmonic cursor, which is much more accurate than a sideband cursor because the members of the
harmonic family are constrained to pass through (and coincide with) zero frequency.

Time-Frequency Analysis:
Signal Decimation Short Time Fourier
Transform (Spectrogram)
• Observe speed varation for the
• Enables Processing a longer whole record.
record • Decide on a relatively small
speed fluctuation part for further
processing
• Enables observing the
general behaviour of the • Observe shaft speed/s and
different gearmesh frequencies
signal and their harmonics

Order Tracking
• Two stages of tracking by phase
demodulating the high speed
shaft 1st harmonic and then the
high speed gearmesh frequency Harmonic/ Sideband
Cursors
• Smearing removed and this
enables using harmonic and
sideband cursors in the next stage

Fig. 2. Data Processing Stages.

 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376 371

4. Results

The Spectrogram (Short Time Fourier Transform) was initially used to examine the speed fluctuation within the gearbox
3rd stage signal as shown in Fig. 3a). A more detailed zoomed-in view (0–700 Hz) is shown in Fig. 3b.
Fig. 3a and b reveal a number of pieces of useful information for the analyst. It gives an indication about the speed
fluctuations and the speed profile during the 600 s of measurement. Note that a reasonably small variation section of the
measured data can be selected between 190 and 290 s.

Fig. 3. (a) Spectrogram for gearbox 3rd stage signal (b) Zoom up to 700 Hz.

Fig. 4. Zoomed spectrogram in the lower frequency range (0–150 Hz).

Fig. 5. FFT of the raw and order tracked signals: (a) Raw signal (0–3500 Hz) (b) Order tracked raw signal (corresponding 0–3500 Hz in orders of the HSS)
(c) Raw (0–500 Hz) (d) Order tracked (corresponding 0–500 Hz in orders of the HSS).
372 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376

Presumed gear mesh frequencies and shaft speeds can be roughly identified. The clearest is the supposed third stage gear
mesh frequency (GMF) appearing just shy of 500 Hz and the second stage gear mesh frequency around 100 Hz.
A further zoom-in in the low frequency region below 150 Hz, as presented in Fig. 4, shows harmonics of the high speed
shaft (HSS). In particular, the 4th harmonic of the HSS is very clear and just slightly above it there appears to be the
intermediate stage GMF.
A section for further processing is selected between the times 190 and 290 s, where speed fluctuation is minimum.
The second stage of processing concentrated on this section for order tracking to remove speed fluctuations. Two stages
of order tracking were applied: the first by demodulating a band around the HSS frequency and the second by demodulating
another band around the now clear third stage gearmesh frequency, which appears to be the 20th harmonic of the HSS.
The results showing the effect of the order tracking process are presented in Fig. 5a and b [0–3500 Hz] and further zoomed
and shown in Fig. 5c and d [0–500 Hz]. The HSS, 3rd stage GMF, 4  HSS, 3  3rd GMF and the 2nd stage GMF appear clearly
as discrete lines. Note the amplitude increase at these frequencies as the energy is now concentrated at the order tracked
frequencies rather than being smeared as before order tracking.
The harmonic and sideband relations on a logarithmic plot (dB scaled) are next plotted and explained to try and identify
the number of teeth on each gear. The first step was to set up a harmonic cursor at the high speed shaft (HSS) and to fine
tune it using the suspected 3rd stage GMF (20  HSS) as presented in Fig. 6. A zoom-in around this 3rd stage GMF was used

Fig. 6. HSS harmonics (A) the first 25th harmonics (0–600 Hz) (b) A zoom in showing harmonics 17–24 (c) A zoom in around the 20  HSS (3rd
stage GMF).

Fig. 7. ISS harmonics (A) the first 120 harmonics (b) A zoom in showing harmonics 106–119 (c) The 25  ISS locks into the suspected 2nd GMF (d) A zoom
in around the 113  ISS (3rd stage GMF).
 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376 373

for an exact tuning of the frequency. A zoom-out was then used to ensure that all the HSS harmonics are well aligned with
the harmonic cursors. Fig. 6a shows the harmonic cursor of the HSS, while Fig. 6b shows the sidebands (also harmonics)
spaced at the HSS around the 3rd stage GMF. A zoom-in around the 20th harmonic (3rd stage GMF) is shown in Fig. 6c to
demonstrate the accuracy of the harmonic cursor. Note that after the order tracking process, the “frequency” scale is actually
an order scale, in this case specified in orders of the high speed shaft (Orders HSS). After the final iteration, the mean output
frequency was found to be 23.60618104 Hz, but the comparisons between shafts should be made on the basis of orders of
one shaft or another.
In Fig. 7, it is shown how the 3rd stage GMF (20  HSS) is in fact the 113th harmonic of a speed around 4.18 Hz. Fig. 7b
shows a family of sidebands around the 3rd stage GMF spaced at the approx. 4.18 Hz (believed to be that of the intermediate
shaft speed (ISS)). It can also be seen from Fig. 7c that the 25th harmonic of this ISS aligns with the suspected 2nd stage
GMF. A zoom-in around the 3rd stage GMF shows how 113  ISS aligns perfectly with the 3rd stage GMF. Note that Figs. 6
and 7 are exactly the same spectrum with the same angular sampling. Zooming-in on the spectrum to adjust the harmonic
cursor was performed to give the necessary alignment and fine tuning for each set of harmonics separately. Tiny
adjustments of the cursor cause some harmonics to lose alignment.
In Table 1, the harmonics based on the two identified fundamental frequencies of the HSS and the ISS obtained from
Figs. 6 and 7, respectively are shown to 11 figure accuracy (expressed as orders of the HSS). The 20  HSS and the 113  ISS

Table 1
HSS harmonics (17–23) and ISS harmonics (105–120).

HSS orders Frequency (Hz) ISS orders Frequency (Hz)

HSS mean fundamental 17 401.30507768 105 438.69893970

frequency (Hz) 23.60618104 18 424.91125872 106 442.87702484
19 448.51743976 107 447.05510998
ISS mean fundamental 20 472.12362080 108 451.23319512
frequency (Hz) 4.17808514 21 495.72980184 109 455.41128026
22 519.33598288 110 459.58936540
23 542.94216392 111 463.76745054
112 467.94553568
113 472.12362082
114 476.30170596
115 480.47979110
116 484.65787624
117 488.83596138
118 493.01404652
119 497.19213166
120 501.37021680

Table 2
Estimated shaft and gear mesh frequencies (for generator mean speed 23.606 Hz).

Estimated frequencies in the gearbox

High shaft speed (Hz) 23.60618104

HS-ST HS-ST GMF
Pinion 20 472.12362080
Gear 113

Intermediate shaft speed 4.17808514

IS-ST IS-ST GMF
Pinion 25 104.45212850
Gear 71

Low shaft speed (sun) 1.47115674

Ring gear teeth 105 Epicyclic mesh

Planet gear 39 31.59643452
Sun gear 27

Output speed (carrier) 0.30091842

Planet pass frequency 0.90275527
Spin frequency  0.81016499

Overall planet stage 4.89

Total ratio 78.447
374 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376

are seen to be the same up to the 11th significant figure indicating an accuracy of better than 1:1010. The number of figures
shows how accurate the harmonic cursor can be after order tracking and gives confidence in the number of teeth for the
pinion and gear of the 3rd stage. The ratio 20  113 is hunting tooth. The closest other harmonics can come is 1/(20  113) ¼
1:2260 (compared with 1:1010).
The numbers of teeth for the second stage pinion (thought to be 25: see Fig. 6c) and matching sun shaft wheel (71) as
well as the planetary stage are shown in Table 2. The bases for filling in the rest of Table 2 are explained using Figs. 7–10.
This parallel stage is also hunting tooth, and 1/(m  n) is 1:1750. Note that because of the possible confusion from choosing a
particular reference “order”, Table 2 uses a “frequency” scale based on the accurate mean frequency of the output shaft of
23.60618104 Hz.
Another series of harmonics was sought for the sun gear shaft, which should also have the 2nd GMF as a member.
In order to do this, the signal was order tracked again, but this time by demodulating a band around the 2nd GMF. Fig. 8
shows the 2nd GMF before and after the order tracking. Order tracking gives a more definite peak.
Fig. 9a–e show the series of harmonics at LSS speed (equivalent to 1.4711567 Hz average frequency) with the 2nd gear
mesh frequency as harmonic No. 71, and including a number of sidebands around the gear mesh frequency. Note also that
there is a pair of strong sidebands spaced at 0.613  LSS (average frequency about 0.902 Hz) (Fig. 9b), with others at 1/3 of
this spacing. These sideband families are however not sub- harmonics of the gear mesh frequency, and are suspected to be
the input shaft speed (believed to be at an average frequency of 0.30 Hz) and its third harmonic the planet pass frequency

Fig. 8. (a) 2nd GMF before being tracked. (b)2nd GMF after order tracking.

Fig. 9. (a) A series of harmonics spaced at LSS (1.4711 Hz), which include the 2nd GMF as harmonic No. 71 (b) Sidebands at 0.902 Hz (planet pass
frequency) (c) A zoom in around the 2nd GMF. (d) A zoom in around the 2  2nd GMF (e) A zoom in around 3  2nd GMF.
 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376 375

Fig. 10. Harmonics of 0.902 Hz (planet pass frequency) Nos. 28, 29, 30, with the middle one also showing 1/3 spacing (0.29 Hz) (b) Harmonics of 0.902 Hz
(planet pass frequency) Nos. 57, 58, 59, with the middle one also showing 1/3 spacing (0.30 Hz).

Fig. 11. Harmonics of 0.3009 Hz, including planetary gear mesh No. 105.

(average frequency about 0.902 Hz). The evidence is thus that the sun gear speed has an average frequency of 1.4711567 Hz.
Note that the HSS is roughly around, but not exactly 16  Input Shaft Speed.
Fig. 10 lends weight to the theory that the input speed is in the vicinity of the 0.30 Hz, indicated from Fig. 9, by showing
some of the harmonics of a series spaced at 0.902 Hz (Nos. 28–30 and 57–59, respectively). Harmonics 29 and 58 have sub-
harmonics at 1/3 of the spacing as well.
It remains to find the details of the gears in the planetary section. These are subject to considerable restraint, as follows.
If S, P, and R are the numbers of teeth on the sun, planet and ring gears, respectively, then in general the following equations
should be satisfied:

R ¼ S þ2P ð1Þ

and either, R and S must be both divisible by 3 (simultaneous meshing with all planets) or the sum of R and S must be
divisible by 3 (offset meshing of the planets). By trial and error, the set of gears indicated in Table 2 were arrived at; viz.
R¼105, P ¼39 and S ¼27. These satisfy the rules given above, and also give approximately the right gear ratio. The overall
ratio of the gearbox with the parameters in Table 2 is 78.447, whereas the actual is thought to be 78.472. It must be admitted
that the hypothetical planetary section is the least likely to be correct, as little evidence could be found to confirm the
details. For example, the planetary gear mesh frequency (at about 31.59 Hz as shown in Table 2) could not be located with
certainty. Fig. 11 shows the harmonics of input speed 0.300 Hz, including the possible planet gear mesh frequency No. 105,
but this is very low. On the other hand, every third harmonic to the left is well represented (some quite strong) and this
sometimes occurs with planetary gears (i.e. the sidebands adjacent to the gear mesh frequency are stronger than the gear
mesh itself) [8].
376 N. Sawalhi, R.B. Randall / Mechanical Systems and Signal Processing 42 (2014) 368–376

Another point is that not all planetary gear sets obey the rules of Eq. (1), as other possibilities can be accommodated by
addendum/dedendum modification. Even so, it is believed that there is strong evidence that at least the parallel section has
been identified correctly.

5. Conclusions

This paper has discussed the case of a blind determination of the numbers of teeth of a wind turbine gearbox with no
speed signal reference, even in a variable speed situation. A four stage approach has been presented. In the first stage, the
acceleration signal measured on the gearbox was decimated to allow analysis of the longer records. The second stage
utilized a Short Time Fourier Transformation (Spectrogram) to observe the speed variation within the measured record and
decide on a part where speed variation is small. Order tracking was used in the third stage to remove speed fluctuations
from the signal and enable the tuning of harmonic/sideband cursors to an extremely high accuracy for a confident analysis.
Finally harmonic/sideband cursors were used to track the different families and sidebands around the gear mesh
frequencies to discover the numbers of teeth on each gear. Results indicated the ability to identify the numbers of teeth
for the two parallel stages with much more confidence than for the planetary stage, where a trial and error approach was
used to decide on the most likely combination for the ring, sun and planet gears.

Acknowledgements

This work is partially supported by Mechanical Solutions Inc. and by Australia's Defence Science and Technology
Organisation (DSTO). Signals used for demonstration purposes were provided by Mechanical Solutions Inc., NJ, USA, and
were obtained under a research project funded by NYSERDA. NYSERDA has not reviewed the information contained herein
and the opinions expressed in this paper do not necessarily reflect those of NYSERDA or the State of New York.

Appendix A. Supplementary material

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.ymssp.
2013.08.017.

References

[1] B.D. Forrester, 1991. Time–Frequency Domain Analysis of Helicopter Transmission Vibration, Department of Defence, Aeronautical Research Laboratory,
Propulsion Report 180.
[2] R.B. Randall, Vibration-based Condition Monitoring: Industrial, Aerospace and Automotive Applications, Wiley, 2011.
[3] J. Urbanek, T. Barszcz, N. Sawalhi, R.B. Randall, Comparison of amplitude-based and phase-based methods for speed tracking in application to wind
turbines, Metrology and Measurement Systems 18 (2) (2011) 295–304.
[4] F. Bonnardot, M. El Badaoui, R.B. Randall, J. Danière, F. Guillet, Use of the acceleration signal of a gearbox in order to perform angular resampling (with
limited speed fluctuation), Mechanical Systems and Signal Processing 19 (2005) 766–785.
[5] F. Combet, L. Gelman, An automated methodology for performing time synchronous averaging of a gearbox signal without speed sensor, Mechanical
Systems and Signal Processing 21 (6) (2007) 2590–2606.
[6] F. Combet, R. Zimroz, A new method for the estimation of the instantaneous speed relative fluctuation in a vibration signal based on the short time
scale transform, Mechanical Systems and Signal Processing 23 (4) (2009) 1382–1397.
[7] M.D. Coats, N. Sawalhi, R.B. Randall, 2009 Extraction of Tacho Information from a Vibration Signal for Improved Synchronous Averaging, Paper
Presented at ACOUSTICS 2009, Australian Acoustical Soc., 23–25 November, Adelaide.
[8] P.D. McFadden, J.D. Smith, An explanation for the asymmetry of the modulation sidebands about the tooth meshing frequency in epicyclic gear
vibration, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science 199 (1) (1985) 65–70.