Anda di halaman 1dari 7

2014 Vibration Institute Annual Training Conference

June 11-13, 2014


San Antonio, TX
FAULT ASSESSMENT OF GENERATOR BEARING FOR A 1.5 MW WIND
TURBINE
Ehsan Mollasalehi, Qiao Sun, David Wood
Department of Mechanical and Manufacturing Engineering
Schulich School of Engineering
University of Calgary
Calgary, AB, Canada
Email: emollasa | qsun | dhwood @ucalgary.ca
Abstract: Demodulation vibration analysis of the generator bearing of a large wind turbine using the
highest energy band calculated by wavelet packet transform is discussed in this paper. Results from the
vibration analysis were consistent with a localized outer-race bearing fault. They were then validated by
cutting the bearing with water jet machine. A significant localized fault was found on the outer race.
Keywords: Condition monitoring; Demodulation; Envelope analysis; Generator bearing; Wind turbine;
Outer race fault
1. INTRODUCTION:
Rolling element bearings have received great attention in the field of condition monitoring. A robust
machinery condition monitoring system is very beneficial to capture a defect so as to prevent machinery
performance malfunctions, or even catastrophic failures by subsequent root cause analysis. Fault detection
can be performed based on information including acoustic emission, stress waveform, oil analysis,
temperature variation, vibration, etc. The most common technique for fault detection is vibration signature
analysis. Vibration monitoring in rotating machineries offers very important information about defects
formed inside the structure of the machine. The information gained by vibration analysis enables the
planning of maintenance. Vibration signature based diagnostics are mainly concerned with the extraction of
those features from a diagnostic signal, which can be related to a good or a defective state of the component.

Figure 1: Typical generator failures from [1]


A survey of over 1000 failed wind turbine generators showed that bearing failure is the dominant cause of
wind turbine generator failure as shown in the Figure 1 [1]. This stresses the need of a special monitoring
system to maintain the system proactively and avoid any huge damage. The online vibration analysis and

proactive maintenance cost a fraction of the total cost of replacement and required man-hours and most
importantly the indirect cost of the turbine shut down.
This paper reflects simple analysis conducted on the operational data received from the generator bearing
of a 1.5 MW wind turbine. The bearing malfunction was identified by ear from the bottom of the tower at
the time of operation.
2. VIBRATION ANALYSIS METHODS:
Vibration analysis is by far the most prevalent method for machine condition monitoring as it has a number
of advantages compared with other methods. One main advantage is that vibration responds without any
intervening time to changes in machine condition. Therefore it can be used for both permanent and
intermittent monitoring. Vibration analysis has superior characteristics for industrial use and is thus chosen
to be the type of analysis in this paper.
Due to rotating elements such as shafts and bearings, vibration signals tend to be periodic. Modulating
frequencies are produced by specific faults of rotating elements including gears, bearing, and shaft. As a
result, the key task to diagnose bearing/gearbox faults is to detect these modulating signals. Demodulation
is an important step in bearing fault detection and thus in the most of machine condition monitoring. The
theory behind this method is well described in the literature such as [2], and is not the purpose of this paper.
2.1 Modulation:
Sidebands occur when a signal is under the effect of modulation, a phenomena that occurs when a so-called
carrier signal, has its amplitude or frequency to vary with time. The first case, when the amplitude varies
with time, is known as amplitude modulation (AM) and the latter case, when the frequency varies with
time, is called frequency modulation (FM) or phase modulation (PM), where the FM simply is the time
derivative of the PM. As the name implies, the carrier frequency carries the intelligence which is called the
modulator. For example in gear vibration signals, the gear mesh frequency and its harmonics are the
carriers and the shaft rotating speeds of the meshing gears are the modulators. Vibration signals with
bearing faults can be modeled as an AM of a carrier signal at the resonance frequency by periodic pulses.
In so-called Envelope Analysis the signal envelope is extracted by amplitude modulation and frequency
analyzed to reveal the repetition frequencies even when these have a small random fluctuation. In the next
section, one of the common approaches that can be used for demodulation is presented.
2.2 Narrowband Demodulation:
Narrowband demodulation techniques select an interesting frequency band for further analysis as an
alternative to analyzing the entire frequency-domain signal. This is performed by plotting the spectrum, and
then select the frequency band in frequency-domain (which usually has the highest energy). Therefore,
bandwidth filter is applied on the FFT of the signal instead of the signal in time-domain.
The algorithm for narrowband amplitude demodulation is now briefly described:
a) Apply the FFT on the input signal.
b) A frequency band of interest is selected.
c) New zero spectrum is generated.
d) The new spectrum is filled with the selected frequency band by shifting it to the left hand end of the new
spectrum, i.e. the lower limit of the selected band starts at the zero frequency.
e) Apply inverse FFT on the spectrum.
f) The narrowband amplitude demodulated signal in time-domain is then calculated by taking the absolute
value of the complex analytic signal.
g) The spectrum of the narrowband amplitude demodulated signal is calculated via absolute value of FFT.
Note that the narrowband selection does not change the length of the signal in the time domain. To have a
better understanding of this procedure, the flowchart of this process is drawn in Figure 2.

Figure 2: Narrowband demodulation technique flowchart [3]


3. VIBRATION RESULTS/DISCUSSIONS:
3.1 Vibration Data:
This section reflects the vibration data analyzed. The data was recorded from the on-line vibration
monitoring systems on the 80 meter 1.5MW wind turbine:
1- Sensor locations:
Drive End Axial (DE-Axi)
Drive End Vertical (DE-Ver)
Non-drive End Vertical (NDE-Ver)
2- Sampling Frequency:
DE-Axi: 5.12 kHz in 3.2 sec, and 25.6 kHz in 0.64 sec
DE-Ver: 0.512 kHz in 16 sec, and 5.12 kHz, 3.2 sec
NDE-Ver: 5.12 kHz in 3.2 sec, and 25.6 kHz, 0.64 sec
3- Averaged rotational speed of the shaft during the data collection period.
4- The operational data were collected in the period of two months.
This paper does not reflect any results from highest and lowest sampling frequency rates. Vibration results
from non-drive end sensor are not reflected here as well. The only set of data shown is associated with 5.12
kHz.
It should be noted that the drive-end bearing was connected to the high speed bearing attached to the gearbox
by a semi-flexible coupling. Both generator bearings were observed to be making noise and scheduled for
replacement.
3.2 Bearing Charactristic Frequencies:
Faults on the inner race, outer race, and rolling elements (balls) of a bearing show peaks, called characteristic
frequencies or orders in the power spectrum of the envelope signal can be calculated and compared with the
peaks in the power spectrum of the envelope signal to identify the source of the bearing faults. In order to
perform the calculations, the following parameters are required [4]:
Diameter of balls: d= 4.5 cm
Pitch diameter: D= 23.5 cm
Contact angle between the ball and the race: = Zero degree

Shaft rotational frequency: f


Number of balls: n=9
Having these values, the following equations are used:
- Ballpass frequency, outer race:
(1)
-

Ballpass frequency, inner race:

Fundamental train frequency (Cage speed):

Ball spin frequency:

(2)
(3)
(

(4)

3.3 Vibration Analysis:


The main purpose of this paper is to show the simple but useful condition monitoring system for generator
bearings which could reduce the thousands of dollars cost of repair or maintenance by early indication of
the problem and scheduling the maintenance on a proper time. For instance, in many cases, the owner of
the wind turbine, in a warm state, replaced a bearing in spring/summer which leads to a higher revenue loss
than winter. This late replacement due to the fault associated with an early indication in vibration signature
could have been completed in winter when Time-of-Use price is usually lower than summer.
To select the bandwidth of interest to perform envelope analysis, wavelet transform was used to select the
highest energy band. Its complete theory and further explanations are discussed in many texts on signal
processing such as [5]. Briefly, wavelet transform breaks down the signal to separate frequency bands,
called detail and approximation. Each approximation and detail then is split into another approximation
and detail, and so on. As the signal energy is proportional to the squared magnitude, then, Root Mean
Square (RMS) values for different bands will be calculated. Considering the sampling frequency of 5120
Hz, the frequency bands are:
0 640 Hz, 640 1280 Hz, 1280 1920 Hz, 1920 2560 Hz
Figure 3 shows the result of DE_Axi signal at the rotational speed of 868.5 rpm. The wavelet
decomposition and RMS showed the highest energy band of 1280 1920 Hz. The primary frequency refers
to 26 Hz which is the ball passing frequency outer race (BPFO) harmonic associated with the other
harmonics.
Figure 4 shows the result of DE_Axi signal at the rotational speed of 1046 rpm. The wavelet decomposition
and RMS calculations showed the highest energy band of 640 1280 Hz. The primary frequency refers to
31.7 Hz which is BPFO harmonic associated with the other harmonics.

Figure 3: DE-Axi at 868.5 RPM


Figure 5 shows the result of DE_Axi signal at the rotational speed of 1181 rpm. The wavelet decomposition
and RMS calculations showed the highest energy band of 1920 2560 Hz. The primary frequency refers to
35.5 Hz which is BPFO harmonic associated with the other harmonics.

Figure 4: DE-Axi at 1046 RPM


Figure 6 shows the envelope analysis of DE_Ver signal at the rotational speed of 1125 rpm. The wavelet
decomposition and RMS calculations showed the highest energy band of 1920 2560 Hz.

Figure 5: DE-Axi at 1181 RPM


However none of the peaks are associated with any of bearing characteristics frequencies. The magnitude is
also much lower than the ones calculated from DE-Axi signals. The other signals of DE-Ver also showed
the same pattern. As it was mentioned earlier, drive end bearing (DE) is close enough to the gearbox to
receive noise (gear mesh signals). Without gearbox configuration information including tooth number of
each gear, supporting bearing models and speed (constant or variable speed), vibration signal from the
bearing cannot be de-noised. This needs further investigation and analysis.

Figure 6: DE-Ver at 1125 RPM

3.4 Visual Inspection:


In the last section, it was shown that there is a high potential of a fault/defect presenting on the outer race.
To validate this conclusion, the bearings disassembled by water jet cutter at the machine shop. As shown in
Figure 7, only the outer race was cut. The bearing was a basic type ceramic insulated bearing, up to 3kV
voltage resistance. The races and balls were made of steel. The cage and the housing were made of brass
and cast iron, respectively.
Water jet cutting machine do not generate any heat, therefore the bearing was not affected and no new
defects/faults were introduced. Figure 8 shows the outer race which has a defect on it. The defect is a
localized-type. Faults in this category maybe produced by excessive load, normal fatigue failure, or
misalignment. On the other hand, the defect in Figure 8 is a burned-out type. Generally, this type of fault
is generated either by electrical discharge, lack of proper lubrication, or shaft/bearing misalignments.
Misalignment which could cause an excessive load on the races/balls, introduced by the shaft or housing,
seems to be more possible. The misalignment produced a very high local load on the area and consequently
high heat. Therefore the lubrication was burned out. The location of the melted area is shown by an arrow
in the left picture in Figure 8 and zoomed in the right picture. There is some evidence of off-track raceways
as well.

Figure 7: Bearing while water jet cutting


This assumption was derived based on the available information and visual inspection. That being said, the
root cause analysis at this stage with the provided vibration data would not yield any firm conclusions.
Root cause analysis requires much complicated analysis with more information. This is the next step which
is vital for any maintenance and scheduling the inspection.

Figure 8: Fault on the outer race

4. CONCLUSION:
This paper briefly demonstrated the application of demodulation analysis on the operational vibration
signals. Only drive-end data from the generator bearing of a wind turbine were chosen to be reflected in the
paper. The results indicated the presence of fault(s) on the outer race before any inspection to be done by
the authors.
To validate the results, the bearing was cut by a water jet cutting machine. Water jet cut was chosen since
no heat is introduced to the rolling elements. It was shown that there was a local fault on the outer race.
Although there was not enough information to perform the root cause analysis, this type of fault most likely
comes from misalignment which can cause very high loads. The misalignment and load could be due to
either the shaft or housing, however it is not certain. Further investigations and inspections are required.
An online condition monitoring system with a reliable analysis could reduce the maintenance cost (cost of
new parts, man-hour at site, loss of production, etc.) and generally the revenue loss by an early indication
of the fault, primarily with a scheduled and proactive maintenance at a proper time. In addition, when a
localized outer race fault has been diagnosed by the monitoring system, for example, the faulty bearing
does not need to be replaced but only to be rotated 180 degrees; the bearing can operate as new to save
cost. The cost of the analysis is small.
It is suggested that collecting additional data such as torque/load will provide supplementary information
regarding the status of the bearing, and will lead to firm results. This information could be either collected
by torque sensors or calculated by having power and rotational speed.
Still, more sophisticated analysis is required to be able to develop a comprehensive analysis tool. For
instance, gearbox data/information will be essential in order to isolate the gear vibration signatures (as
shown in Figure 6), so that the monitoring system can handle all typical mechanical faults,.
5. ACKNOWLEDGMENT:
The authors would like to thank the TransAlta Corporation for providing the operational data and the
bearings, and the Natural Sciences and Engineering Research Council (NSERC) of Canada for financial
support under the Industrial Research Chair program.

6. REFERENCES:
[1] Alewine, K., Chen, W., Wind Turbine Generator Failure Modes Analysis and Occurrence,
Windpower 2010, Dallas, Texas, May 24-26, 2010
[2] Rabe, U., K. Janser, and Wt Arnold. "Vibrations of free and surface-coupled atomic force microscope
cantilevers: theory and experiment", Review of Scientific Instruments 67.9 (1996): 3281-3293.
[3] Andreas Meisingseth, Demodulation Techniques in Gearbox Diagnostics, July 2012
[4] Randall, R. Vibration Based Condition Monitoring, 2012, John Wiley & Sons]
[5] Chui, C. K., 1992, An Introduction to Wavelets, Academic Press, New York, pp. 1-200.
7. BIOGRAPHY:
Ehsan Mollasalehi received the B.Sc. degree in Mechanical Engineering (M.E.) from
K.N. Toosi University of Technology, Tehran, Iran in 2010, and the M.Sc. degree in M.E.
from the University of Calgary, Calgary, Canada in 2012. He is currently a Ph.D. student
in M.E. at the University of Calgary.
His primary research interests are in the areas of fault detection and diagnostics, vibration
and noise control, and Dynamics. Currently, he is working on condition monitoring of
mechanical components of large scale wind turbines.
He is a member of ASME, and Canadian Machinery Vibration Association (CMVA).