Annual Conference of the Prognostics and Health Management Society, 2011

Statistical vibration analysis for predictive maintenance of

machines working under large variation of speed and load

Luisa F. Villa1, Aníbal Reñones1, Jose R. Perán1 and Luis J. de Miguel2

1
CARTIF Foundation, Parque Tecnológico de Boecillo, 47151 Boecillo, Valladolid, Spain
luivil@cartif.es
aniren@cartif.es
peran@eis.uva.es
2
University of Valladolid, Valladolid, Spain
luimig@eis.uva.es

ABSTRACT 1. INTRODUCTION
Prognosis of defects for machines working under large Vibration analysis has been implemented and studied in
variation of speed and load conditions is a topic still rotating machinery for many years, and it is widely
under development. Wind turbines are recent examples accepted as one of the main techniques for condition
of such kind of machines that need reliable diagnosis based maintenance (Hameed, 2009). With the advance
methods. Vibration analysis can be of very limited use of technology, more complex machines that operate
when the speed variation is too high. An effective under more severe conditions have been developed; an
angular resampling method can be very valuable as the example of these conditions are those who operate
first step of vibration signal processing but it is under varying loads and speeds like wind turbines,
important to know what are the appropriate variables to excavators and helicopters (Barszcz, 2009; Blunt, 2006;
be monitored. Combet, 2009; Bartelmus, 2009). In these kinds of
machines, gear transmissions play a crucial role in
The authors present a statistical analysis method terms of their reliability.
consisting of a linear model based on the parameters
that characterize the system, in our case the variable The initial research in the area of transmission damage
speed and load, and the fault condition to which the detection was focused on vibration signal analysis
system is subjected. With this method can be (Davies, 1998). At first, as discussed in (Samuel, 2005),
determined if the variable analyzed is significant, that is the statistical characteristics of the signal in the time
to say if are sensitive to these parameters and hence can domain were the primary focus of study. However, the
detect the fault faster. The aim of implementing this field quickly expanded to include spectral analysis,
method is to reduce the number of variables to be time-frequency analysis, wavelet analysis, neural
monitored, resulting in a savings not only in measuring networks and mathematical modeling. This field is
equipment but also in times of processing and continuing to grow. As new signal processing
analyzing information. techniques emerge, they are applied to the transmission
damage detection problem and must be accommodated
The results of vibration analysis of a test-bed working to the needs and specificities of each mechanical
under large variation of speed and load are shown. system.
Different tests with increasing level of defects are tried
and the corresponding vibration is analyzed and For systems which work under variable speed and load
modeled so an effective detection and prognosis can be conditions one of the most appropriate signal
done. Taking in to account such variation of speed and processing method is angular resampling, however,
load for the vibration modeling can lead to a very using this method is only the fisrt step because it is
sensitive detection of incipient defects.* necessary to analyze the information obtained and it is
also important to determine that the variables that are
being analyzed are those that provide the best
* information on fault diagnosis. For the selection of
This is an open-access article distributed under the terms of
the Creative Commons Attribution 3.0 United States License,
these variables, the authors present in this paper a
which permits unrestricted use, distribution, and reproduction statistical method based on a linear model to determine
in any medium, provided the original author and source are the sensitivity of each variable with respect to the
credited. failure and system operating conditions.

1
 Annual Conference of the Prognostics and Health Management Society, 2011

The main motivation for this work is the diagnosis of For the above reasons, the vibration monitoring
wind turbines, so the test bed that will be shown, is systems require signal processing procedures to
used to approach the phenomenon and failures of a real compensate for the fluctuations in shaft speed and the
wind turbine. amplitude modulation caused by the variable load
(UpWind, 2006; Stander, 2002; Stander, 2005).

2. VARIABLE SPEED AND LOAD In the fluctuating load research many studies have been
RESERARCH developed using various methods of signal analysis,
among these studies are those by Stander, Heyns, Zhan
For the analysis of systems that operate under variable and Bartelmus (Stander, 2006; Zhan, 2006; Bartelmus,
speed conditions, efforts were made to find techniques 2009). However, until now no studies have been
that allow for better processing and analysis of the developed in an extended work range of speed and
signals captured from these systems. A theory of load, and that is where our research focuses.
interpolation applied to time domain averaging was
presented in (McFadden, 1989; McFadden, 1991) as an
alternative to averaging when there is no rotational 3. ANGULAR RESAMPLING ALGORITHM
reference signal.
FOR LARGE SPEED VARIATION
Due to the fact that many systems are working under Due to the speed variations caused by the operating
variable speed, the technique known as order tracking conditions itself, and load variations, as is the case of
is very common in vibration analysis. Each of the steps wind turbines, it is necessary to process the vibration so
that comprise the order tracking method are explained its frequency content can be analyzed. The angular
in (Fyfe, 1997). In this method the main algorithm is resampling technique can be used to solve this problem.
the angular resampling, which repositions the samples
of vibration to be equivalents to a signal measured at The works that have been developed previously on the
constant speed. In this article comparisons are made at issue of angular resampling are applicable to cases in
each step and the best alternatives for each are given. which the speed fluctuations are small (Bonnardot,
Later, (Bossley, 1999) presented a hybrid, computed 2005; McFadden, 1989; Fyfe, 1997; Bossley, 1999).
order tracking method to perform angular resampling The application of this kind of signal processing to the
which was compared with two previously proposed vibration analysis of machines like wind turbines is
methods by comparing the results. limited, because the angular speed and hence
acceleration variations experienced in a wind turbine
A method to perform angular resampling was presented are high and are not predictable as they depend on the
in (Bonnardot, 2005) using the acceleration signals wind.
directly without the need for an encoder signal, but this
method has the limitation that it can only be used when The method of angular resampling algorithm proposed
speed variations are small and it requires a sufficient by (Fyfe, 1997) is valid for linear profile of speed. The
number of harmonics. authors in (Villa, 2011) presented an evolution of this
method, an angular resampling algorithm for a general
An extension to the algorithm proposed by Bonnardot case of variable speed and a generic number of
was presented by Combet. However, it is not advisable keyphasors.
for use in the case of very large speed variations, such
as during acceleration (Combet, 2007; Combet, 2009). The method proposed by (Fyfe, 1997) includes the
following steps; first records the data at constant ∆t
Another factor affecting the variability of the speed is increments, and them resamples this signal to provide
the fluctuating load that modulates the amplitude of the the desired data at constant ∆θ increments, based on a
vibration signal measurement and causes changes in the
keyphasor signal. To determine the resample times, it is
rotational speed of the system. Changes in the system
assumed that the shaft is undergoing constant angular
speed cause a modulation in the characteristic
frequencies of the mechanical elements of the system acceleration, the shaft angle; θ is described by the
(Stander, 2005). following quadratic equation:

The resulting spectra of a signal under fluctuating load θ (t ) = b0 + b1 t + b2 t 2 (1)

conditions presents multiple peaks known as
"smearing” in the region around the characteristic
frequencies of the system. The unknown coefficients b0 , b1 and b2 are found by
fitting three successive keyphasor arrival times

2
 Annual Conference of the Prognostics and Health Management Society, 2011

( t1 , t 2 and t 3 ), which occur at known shaft angle 4. TEST-BED TESTS

increments. Once the resample times are calculated, the The experiments presented in the next sections are to
corresponding amplitudes of the signal are calculated simulate in a test bed the behavior that occurs in the
by interpolating between the sampled data. After the wind turbines. This test bed is used to simulate
amplitudes are determined, the resampled data are different defects under variable load and speed and it is
transformed from the angle domain to the order domain controlled.
by means of an FFT.
The right side of the test bed (figure 3) is composed of
Due to the fact that the speed varies considerably a motor (instead of the generator of a wind turbine), a
between consecutive rotations, the decision taken was parallel gearbox and a planetary gearbox. Both
to use more than one pulse per revolution, and to use gearboxes resemble the configuration and the gear ratio
the full profile of the speed instead the analysis of three of a commercial wind turbine of 1:61.
consecutive samples like the algorithm presented in
(Fyfe, 1997). The main improvement of the proposed To simulate the load variable that is under the drive
algorithm in (Villa, 2011) is to take full advantage of train due to the randomness of the wind, an electrical
the whole measured speed to obtain an accurate angular brake has been added to the test bed. For coupling it
resampled vibration. with the right side of the test bed, it is necessary to add
another gearbox to compensate for the torque in the
Figure 1 shows the results of this algorithm applied to a slow axis against the torque of the motor in the fast
simulated signal created with parameters that shaft.
characterize the test bed described in the next section.
The simulated signal was created with a wide range of
speed variation of and our resampling algorithm can
process it without any problem.

Figure 3. Test-bed
Figure 1. Spectrum before and after angular resampling
for a simulated signal
For the acquisition of vibration signals we used four
accelerometers distributed in axial and radial position
The angular resampling method developed by the in the gearboxes located on the right side of the test-
authors (Villa, 2011) was also tested with experimental bed.
data with satisfactory results. The data shown
corresponds to a bearing with damage in the inner ring
of the fast Shaft (figure 2). 5. EXPERIMENTATION
The failures simulated on the test bed were unbalance
and misalignment, starting with small values of defects
and increasing with each new set of measurements to
simulate a progressive failure (table 1). The table shows
the value of the weight in grams and the equivalent
percentage of the total weight of the rotor test bed, and
the thickness of the sheet used to misalign and their
respective angle of misalignment.

To guarantee variable speed and load conditions,

Figure 2. Spectrum before and after angular resampling different profiles were generated to cover a random
for an experimental signal range of velocity between 1000 and 1800 rpm, and a
range between 0 and 100% of load. An example of
these profiles are shown in figure 4.

3
 Annual Conference of the Prognostics and Health Management Society, 2011

Table 1: Kind of faults Table 2. Variables monitored

Unbalance Misalignment Type of variable Number of
gr % mm º variables
Unbalance 5.79 0.077 Misalignment 0.75 1.53 Statistical variables 5
A A Electrical variables 4
Unbalance 9.13 0.12 Misalignment 2 0.78 Order 1X to 10X 10
B B Gears 12
Unbalance 19.5 0.26 Common orders for the 4 accelerometers 58
C Common frequency bands for the 4 40
Unbalance 28.8 0.38 accelerometers
D Planetary Axial 15
Planetary Radial 4
Parallel Axial 10
Parallel Radial 8

6. ANALYSIS AND RESULTS

From monitored variables, we selected the classical
parameters used for the analysis of the defects that are
in analysis; the harmonic 1x in radial direction of the
planetary gearbox to analyze the unbalance, and the
harmonics 1x and 2x in the radial and axial direction of
the planetary gearbox to analyze misalignment.
Figure 4. Speed and load profile

These profiles were generated to cover a full day of 6.1 Unbalance analysis
measurements (24h), with constant intervals of speed
and load every 100 sec. Speed measurements were First of all an analysis of the radial vibration level (first
generated starting at 1000 rpm because this is the harmonic) for the different defects was made. It is
approximately equivalent speed in the slow axis when a presented the vibration for the 4 levels of unbalance
wind turbine begins to generate energy. compared with the vibration without defect, and can be
clearly seen that the defects cannot be distinguished
Captures were made of 72 seconds with each of the directly (figure 5) and are not statistically different
four accelerometers mentioned in section 4 with a (figure 6) due to the fact that the amount of unbalance
sampling frequency of 25600 Hz. The speed signal and the speed shaft are very low.
captured in the slow axis used for angular resampling
was sampled at a frequency of 6400 Hz.

The variables monitored from the vibration signals

captured are order and natural frequencies of the
system determined experimentally, calculated orders
for specific elements such as gears and bearings,
statistical parameters extracted from the time domain
and the harmonics of order 1 to 10 of the rotation speed
resulting in a final set of 166 variables (table 2).

Figure 5. Vibration level harmonic 1X planetary radial

(unbalance)

4
 Annual Conference of the Prognostics and Health Management Society, 2011

It must be said that for the case of unbalance, due to the

small amount of added mass that can be practically
Harmonic 1x Planetary Radial
used, the slow speed (below 0.5 Hz) and the wide range
of working conditions (speed and load) the classical
0.00045

parameter used, that is the radial vibration at the first

order, is not statistically conclusive. Therefore it is
necessary to rely on other parameters like for example
the 6x order. Although the defects cannot be
0.00035

distinguished directly (figure 7) with this parameter and

are not statistically different (figure 8), the default
Vibration (g)

parameter is significant if the vibration of 6x order is

modeled using the speed, load and defect level (table
0.00025

4).
0.00015

0 5.79 9.13 19.5 28.8

Defect level (gr)

Figure 6. Box-plot harmonic 1X planetary radial

(unbalance)

If the same data (radial vibration level of first

harmonic) is modeled with a linear model that takes
into account the speed, load and the unbalance level the
resulting parameters are not significant (table 3).

The equation of the model is

Y = (b0 + b3 F ) + b1 S + b2 L (2)
Figure 7. Vibration level harmonic 6X planetary radial
(unbalance)
Where Y is the vibration dependent on the speed, load
and defect, b0 + b3 F is the intercept, b1 S is the
slope as a function of speed and b2 L is the slope as a
function of load.

Table 3: Coefficients modeling harmonic 1X planetary

radial (unbalance)
Estimate Std. Error t value Pr(>|t|)
b0 2.202e-04 4.208e-06 52.322 < 2e-16 ***
b1 8.313e-05 9.484e-06 8.766 < 2e-16 ***
b2 -1.895e-05 2.262e-06 -8.377 < 2e-16 ***
F=5.79 -2.645e-07 1.819e-06 -0.145 0.88438

F=9.13 4.686e-06 1.814e-06 2.583 0.00982 **

F=19.5 -8.896e-07 1.824e-06 -0.488 0.62582
F=28.8 2.527e-06 1.810e-06 1.396 0.16274

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

5
 Annual Conference of the Prognostics and Health Management Society, 2011

Table 5. Selected variables to be monitored for

Harmonic 6x Planetary Radial
unbalance
Type of variable Number of
1e-04

variables
Statistical variables 1
Electrical variables 0
Order 1X to 10X 10
8e-05

Gears 5
Vibration (g)

Common orders for the 4 accelerometers 17

Common frequency bands for the 4 9
6e-05

accelerometers
Planetary Axial 0
Planetary Radial 1
4e-05

Parallel Axial 0
Parallel Radial 2
2e-05

0 5.79 9.13 19.5 28.8

6.2 Misalignment analysis
Defect level (gr)
In the case of misalignment, the results obtained from
Figure 8. Box-plot harmonic 6X planetary radial the visual and the statistical analyses are similar to the
(unbalance) unbalance, that is, with the analysis of the vibration
level alone cannot be differentiated statistically the
amount of misalignment of the harmonic 1x axial
Table 4: Coefficients modeling harmonic 6X planetary (Figures 9 and 10) and radial (Figures 11 and 12) or 2x
radial (unbalance) axial (Figures 13 and 14) and radial (Figures 15 and
Estimate Std. Error t value Pr(>|t|) 16).
b0 2.159e-05 4.378e-07 49.31 <2e-16 ***
b1 2.902e-05 9.866e-07 29.42 <2e-16 *** Harmonic 1x Planetary Axial

b2 -5.538e-05 2.353e-07 -235.31 <2e-16 ***

8e-04

F=5.79 -2.987e-06 1.892e-07 -15.79 <2e-16 ***

7e-04

F=9.13 -5.409e-06 1.887e-07 -28.66 <2e-16 ***

F=19.5 -2.864e-06 1.898e-07 -15.09 <2e-16 ***
6e-04

F=28.8 -2.940e-06 1.883e-07 -15.61 <2e-16 ***

Vibration (g)

5e-04

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
4e-04

With the modeling parameters of speed, load and defect

can be determined which variables are most sensitive to
failure and they are to be monitored. Based on this
3e-04

analysis, a selection of the most significant variables

for the detection of the faults has been done resulting in
2e-04

a final set of 45 variables (table 5).

0 0.75 2

Defect level (mm)

Figure 9. Box-plot harmonic 1X planetary axial

(misalignment)

6
 Annual Conference of the Prognostics and Health Management Society, 2011

Figure 10. Vibration level harmonic 1X planetary axial Figure 12. Vibration level harmonic 1X planetary radial
(misalignment) (misalignment)

Harmonic 1x Planetary Radial Harmonic 2x Planetary Axial

0.00045

4e-04
0.00035

3e-04
Vibration (g)

Vibration (g)
0.00025

2e-04
0.00015

1e-04

0 0.75 2 0 0.75 2

Defect level (mm) Defect level (mm)

Figure 11. Box-plot harmonic 1X planetary radial Figure 13. Box-plot harmonic 2X planetary axial
(misalignment) (misalignment)

7
 Annual Conference of the Prognostics and Health Management Society, 2011

Figure 14. Vibration level harmonic 2X planetary axial Figure 16. Vibration level harmonic 2X planetary radial
(misalignment) (misalignment)

On the other hand, a linear model of the vibration using

Harmonic 2x Planetary Radial
the speed, load and defect levels as independent
variables can lead to a detection and separation of the
defect levels, where the significance level for the
0.00012

different defect levels are statistically significative

(small p-value) as is shown for the harmonic 1x axial
0.00010

(table 6) and radial (table 7) or 2x axial (table 8) and

radial (table 9).
Vibration (g)

0.00008

Table 6: Coefficients modeling harmonic 1X planetary

axial (misalignment)
0.00006

Estimate Std. Error t value Pr(>|t|)

b0 3.618e-04 1.267e-05 28.568 <2e-16 ***
0.00004

b1 3.281e-04 2.877e-05 11.405 <2e-16 ***

b2 -2.358e-05 6.661e-06 -3.539 0.000408 ***

0 0.75 2
F=0.75 -5.325e-05 4.190e-06 -12.709 <2e-16 ***

Defect level (mm) F=2 -5.301e-05 4.224e-06 -12.550 <2e-16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Figure 15. Box-plot harmonic 2X planetary radial
(misalignment)
Table 7: Coefficients modeling harmonic 2X planetary
axial (misalignment)
Estimate Std. Error t value Pr(>|t|)
b0 2.713e-04 3.826e-06 70.923 <2e-16 ***
b1 -2.378e-04 8.690e-06 -27.365 <2e-16 ***
b2 -4.248e-06 2.012e-06 -2.112 0.0348 *
F=0.75 -2.818e-05 1.266e-06 -22.264 <2e-16 ***
F=2 -3.300e-05 1.276e-06 -25.867 <2e-16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

8
 Annual Conference of the Prognostics and Health Management Society, 2011

Table 8: Coefficients modeling harmonic 1X planetary 7. CONCLUSION

radial (misalignment)
Estimate Std. Error t value Pr(>|t|) The present work shows the first results of a statistical
analysis of vibration applied to a test-bed working
b0 2.436e-04 5.262e-06 46.293 <2e-16 *** under large variability of speed and load conditions. To
b1 1.833e-05 1.195e-05 1.533 0.125 solve the problems caused by variations in speed and
load, the vibration signals are processed with an
b2 -2.544e-05 2.767e-06 -9.195 <2e-16 ***
angular resampling method developed by the authors.
F=0.75 -3.829e-05 1.741e-06 -21.995 <2e-16 *** Based on this signal processing technique the classical
F=2 -2.943e-05 1.755e-06 -16.769 <2e-16 ***
vibration order parameters can be used to detect
incipient faults like small unbalance and
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 misalignments. The authors show how through a
statistical vibration analysis taking into account the full
working range of the test-bed (speed and load) as
Table 9: Coefficients modeling harmonic 2X planetary independent parameters can be determined which
radial (misalignment) variables are most sensitive to these parameters and to
Estimate Std. Error t value Pr(>|t|) the failure and can be reduced the number of variables
b0 9.265e-05 1.710e-06 54.185 <2e-16 *** to be analyzed do not always coincide with the
variables recommended by the theory of vibration
b1 -6.068e-05 3.884e-06 -15.623 <2e-16 *** analysis.
b2 -1.122e-05 8.992e-07 -12.474 <2e-16 ***
At present, the authors are working on an automated
F=0.75 4.753e-06 5.657e-07 8.402 <2e-16 ***
selection of parameters using computer intelligence
F=2 -3.239e-06 5.703e-07 -5.680 1.49e-08 *** methods, and are also working on a prognosis based on
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 statistical significance levels of different models.
Another important defect simulation under work is gear
failure. The aim of the authors is to implement the
Because cases of misalignment used in these tests are prognosis algorithms for the detection of mechanical
very severe, it is easy to detect with all the variables, defects in the drive train of wind turbines.
however the method presented in this section to
determine the most sensitive has reduced the number of
variables to 133, in Table 10 shows this result. As the ACKNOWLEDGMENT
classical parameters used for monitoring misalignment, The research work developed in this paper was partially
orders 1X and 2X in radial and axial directions are supported by the funded project: CICYT reference
among this group of 133 variables, these are the DPI2009-14608-C02-02 “Diagnosis of wind turbines
variables to be monitored for this type of failure. based on analytical redundancy”.

Table 10. Selected variables to be monitored for Special thanks to Roberto Arnanz for his work in the
misalignment data recording and processing during the project.
Type of variable Number of
variables REFERENCES
Statistical variables 5
Electrical variables 0 Z. Hameed, Y. S. Hong, Y. M. Cho, S. H. Ahn, C. K.
Song. (2009). Condition monitoring and fault detection
Order 1X to 10X 10
of wind turbines and related algorithms: a review,
Gears 12
Renewable and Sustainable Energy Reviews, vol. 13,
Common orders for the 4 accelerometers 49 pp 1-39.
Common frequency bands for the 4 32
accelerometers T. Barszcz, R. B. Randall. (2009). Application of
Planetary Axial 11 spectral kurtosis for detection of a tooth crack in the
Planetary Radial 2 planetary gear of a wind turbine, Mechanical systems
Parallel Axial 7 and signal processing, vol. 23, pp 1352-1365.
Parallel Radial 5
D. M. Blunt, J. A. Keller. (2006). Detection of a fatigue
crack in a UH-60A planet gear carrier using vibration
analysis, Mechanical systems and signal processing,
vol. 20, pp 2095-2111.

9
 Annual Conference of the Prognostics and Health Management Society, 2011

F. Combet, R. Zimroz. (2009). A new method for the operation under fluctuating load conditions,
estimation of the instantaneous speed relative Mechanical systems and signal processing, vol. 16 pp
fluctuation in a vibration signal based on the short time 1005-1024.
scale transform, Mechanical Systems and Signal
Processing, vol. 23, pp 1382-1397. C. J. Stander, P. S. Heyns. (2006). Transmission path
phase compensation for gear monitoring under
W. Bartelmus, R. Zimroz. (2009). Vibration condition fluctuating load conditions, Mechanical systems and
monitoring of planetary gearbox under varying external signal processing, vol. 20 pp 1511-1522.
load, Mechanical systems and signal processing, vol.
23, pp 246-257. Y. Zhan, V. Makis, A. K. S Jardine. (2006). Adaptive
state detection of gearboxes under varying load
A. Davies. (1998). Handbook of condition monitoring conditions based on parametric modeling, Mechanical
techniques and methodology, Chapman and Hall, UK. systems and signal processing, vol. 20 pp 188-221.

P. D. Samuel, D. J. Pines. (2005). A review of L. F. Villa, A. Reñones, J. R. Perán, L. J. Miguel.

vibration-based techniques for helicopter transmission (2011). Angular resampling for vibration analysis in
diagnostics, Journal of Sound and Vibration, vol. 282, wind turbines under non-linear speed fluctuation,
pp 475-508. Mechanical systems and signal processing, Article in
Press.
P.D. McFadden. (1989). Interpolation techniques for
time domain averaging of gear vibration, Mechanical
Systems and Signal Processing, vol. 3 pp 87-97.

P.D. McFadden. (1991). A technique for calculating the

time domain averages of the vibration of the individual
planet gears and the sun gear in an epicyclic gearbox,
Journal of Sound and Vibration, vol. 144 pp 163-172.

K.R. Fyfe, D.S. Munck. (1997). Analysis of computed

order tracking, Mechanical Systems and Signal
Processing, vol. 11 pp 187-205.

K.M. Bossley, R.J. Mckendrick, C.J. Harris, C. Mercer.

(1999). Hybrid computed order tracking, Mechanical
Systems and Signal Processing, vol. 13 pp 627-641.

F. Bonnardot, M. El Badaoui, R. B. Randall, J. Danière,

F. Guillet. (2005). Use of the acceleration signal of a
gearbox in order to perform angular resampling (with
limited speed fluctuation), Mechanical Systems and
Signal Processing, vol. 19 pp 766-785.

F. Combet, L. Gelman. (2007). An automated

methodology for performing time synchronous
averaging of a gearbox signal without speed sensor,
Mechanical Systems and Signal Processing, vol. 21 pp
2590-2606.

C. J. Stander, P. S. Heyns. (2005). Instantaneous

angular speed monitoring of gearboxes under non-
cyclic stationary load conditions, Mechanical systems
and signal processing, vol. 19 pp 817-835.

UpWind. (2006). “state of the art” report condition

monitoring for wind turbines.

J. Stander, P. S. Heyns, W. Schoombie. (2002). Using

vibration monitoring for local fault detection on gears

10