machines

Article
Full-Scale Wind Turbine Vibration Signature Analysis
Xavier Escaler 1, * and Toufik Mebarki 2
1 Center for Industrial Diagnostics and Fluid Dynamics (CDIF), Universitat Politècnica de Catalunya,
Av. Diagonal 647, 08028 Barcelona, Spain
2 Schaeffler Iberia S.L.U., Foment 2, 08960 Sant Just Desvern, Spain; [email protected]
* Correspondence: [email protected]; Tel.: +34-934-012-599




Received: 7 November 2018; Accepted: 5 December 2018; Published: 7 December 2018


Abstract: A sample of healthy wind turbines from the same wind farm with identical sizes and designs
was investigated to determine the average vibrational signatures of the drive train components
during normal operation. The units were variable-speed machines with three blades. The rotor
was supported by two bearings, and the drive train connected to an intermediate three-stage
planetary/helical gearbox. The nominal 2 MW output power was regulated using blade pitch
adjustment. Vibrations were measured in exactly the same positions using the same type of sensors
over a six-month period covering the entire range of operating conditions. The data set was
preliminary validated to remove outliers based on the theoretical power curves. The most relevant
frequency peaks in the rotor, gearbox, and generator vibrations were detected and identified based on
averaged power spectra. The amplitudes of the peaks induced by a common source of excitation were
compared in different measurement positions. A wind speed dependency of broadband vibration
amplitudes was also observed. Finally, a fault detection case is presented showing the change of
vibration signature induced by a damage in the gearbox.

Keywords: wind turbine; condition monitoring; vibrations; frequency content; drive train; tower

1. Introduction
The rapid development and growth of the wind energy industry has resulted in the installation of
a significant fleet of onshore and offshore wind turbines across the world. The current promotion of
wind electricity seems to indicate that more wind farms will be required in the near future, comprising
units with larger power outputs that are mainly located offshore. Therefore, wind farm profitability
must be increased by ensuring that wind turbines (WT) operate at a high capacity. Thus, proper
operation and maintenance (O&M) is critical for maximizing the returns on the wind investments and
for optimizing the total cost of ownership (TCO). This strategy seeks to minimize the production costs
per unit of energy generated and to improve the turbine performance [1].
In this sense, condition monitoring (CM) systems have been developed to detect anomalies with
the goal of minimizing machine downtime and maximizing availability. The early detection of faults
is crucial for performing predictive (condition-based) maintenance on units in a wind farm [2–6].
Nevertheless, the complexity of the WTs, which are subjected to variable loads and speeds and are
made with flexible components, such as the blades [7], challenges the effectiveness of such systems.
Among various methods [8], assessing a machine’s condition based on its measured vibrations is one
of the most reliable techniques, as the vibration levels of the wind turbines significantly influence the
stress and fatigue of the components. This method has been extensively utilized in rotating machines,
producing effective results in various types of industrial machines. However, the traditional processes
and procedures that were developed in other industries, such as the oil and gas, hydropower, or
nuclear power industries, are not valid for the wind industry, as the machine design, operation, and
environment differ significantly.

Machines 2018, 6, 63; doi:10.3390/machines6040063 www.mdpi.com/journal/machines

Machines 2018, 6, 63 2 of 16

Initially, relevant operation parameters were continuously monitored using the integrated
supervisory control and data acquisition (SCADA) system [9,10]. Then, acquiring and measuring
vibrations at high frequency rates using dedicated systems was found to be more reliable for evaluating
the states of the various components of the drive train and the nacelle/tower system [11–13]. In part,
this method was utilized due to the multiple drive train rotating speeds and the rapid evolution of
their incipient failures. Recently, new signal processing methods have been developed and tested to
overcome the difficulty of applying traditional vibration analyses [14–16], largely because WTs are
submitted to complex sources of vibration. These excitations are due to wind loads on the rotor blades
and the tower, the inertial forces of the rotating parts, the natural frequencies of various components,
the mechanical forces in the power transmission system (including gear meshing processes) and the
electromagnetic forces acting on the generator. Moreover, the rotor shaft turns at low speeds, the wind
loads are stochastic, and the regulation system also induces transient conditions and unsteadiness.
To establish a CM system that is based on vibrations, the only reference that is currently available is
the Association of German Engineers (VDI) standard VDI 3834 [17]. This standard provides criteria and
recommendations regarding the measurement and evaluation of the mechanical vibrations of onshore
wind energy turbines with gears and components. However, the suggested values for evaluating
the vibration levels are not suitable for the early detection of faults. Thus, if threshold values are
exceeded, specific causes cannot be identified in detail. The vibratory behavior of a turbine depends
on the manufacturer, design, and installation type. In all cases, it is imperative to account for the entire
wind turbine, including the drive train components, rotor blades, nacelle, and tower. The resulting
vibration behavior depends on the individual response of each part, the interactions throughout the
entire wind energy installation, the resonance conditions, the types and ranges of excitation or loads,
and the operating scenario.
The main objective of this study is to determine whether the vibration content of the drive train
components measured during steady conditions is repetitive and if it follows a defined pattern that is
based on the entire range of operating conditions. This objective is achieved by comparing WTs with
the same design and characteristics. These WTs have been monitored in exactly the same positions
using the same type of CM system over a sufficiently long period of time. The analysis steps are
as follows:

1. detect and identify the most relevant frequency peaks using the median power spectrum at each
measurement position based on the particular drive train kinematics;
2. determine whether differences exist between the measurement points and assess the sensitivity
to different sources of excitation calculating the mean peak amplitudes and their deviations; and,
3. establish the dependency of the vibration amplitudes on the wind speed.

The identification of the WT vibration signature is necessary to guarantee a successful CM system.

A well-designed WT in good condition will exhibit a certain level of vibration depending upon the
machine design and operating conditions. This information must be used as a reference to easily detect
the occurrence of new frequency peaks or variations in existing frequencies. Frequency variations
indicate changes in the machine condition due to incipient failures or increased component wear.
Trend analyses of the narrow band frequencies around these peaks can be used to manage maintenance
strategies and minimize or eliminate problems.

2. Machines and Measurement System Description

2.1. Wind Farm Configuration

The studied WTs are located at the same onshore wind farm. Of the 20 total WTs, we selected
seven machines with the same speed ratio and manufacturer (numbers 1, 2, 4, 7, 11, 13, and 18). Their
relative locations in the farm are illustrated in Figure 1, which shows the units from a topographic
view. The minimum distance between any pair of units is more than 200 m.
 Machines 2018, 6, x FOR PEER REVIEW 3 of 16

relative
Machines 2018,locations
6, 63 in the farm are illustrated in Figure 1, which shows the units from a topographic
3 of 16
view. The minimum distance between any pair of units is more than 200 m.

Figure
Figure 1. 1. Topographicdistribution
Topographic distributionof
of the
the wind
wind turbines
turbines in
in the
the wind
windfarm.
farm.Investigated units
Investigated areare
units
denoted using a red circle.
denoted using a red circle.

2.2.2.2.
WindWind Turbine
Turbine Specifications
Specifications
Each
Each WT WT consists
consists ofofa agenerator
generatorcoupled
coupled to
to aa rotor
rotor via
via aagearbox.
gearbox.The
Theconical
conicaltubular
tubularstyle tower
style tower
is constructed
is constructed from
from steel,while
steel, whilethe
thefoundation
foundation isis made
made of
of concrete.
concrete.The
Therotor
rotorshaft is is
shaft supported
supported by by
two bearings separated from the gearbox. All of the wind turbines use the same
two bearings separated from the gearbox. All of the wind turbines use the same 50 Hz, four-pole 50 Hz, four-pole
generator
generator driven
driven byby a three-stageplanetary-helical
a three-stage planetary-helical gearbox
gearbox(gearbox
(gearboxwith
withone
oneplanetary
planetary stage andand
stage
two additional parallel-shaft stages). Four different gearbox models are installed among the units. In
two additional parallel-shaft stages). Four different gearbox models are installed among the units.
addition, the turbines have horizontal-axis rotors with three blades, totaling 90 m in diameter. The
In addition, the turbines have horizontal-axis rotors with three blades, totaling 90 m in diameter. The
WTs are variable-speed machines that use rotor blade adjustment pitch to regulate the output power.
WTs are variable-speed machines that use rotor blade adjustment pitch to regulate the output power.
The power per unit is 2000 kW for a rated rotor speed of 14.5 rpm and a rated wind speed of 12 m/s.
TheThepower per unit is 2000 kW for a rated rotor speed of 14.5 rpm and a rated wind speed of 12 m/s.
cut-in and cut-out wind speeds are 4 and 25 m/s, respectively. Therefore, the rotor rotational
Thespeed
cut-in and cut-out
can range from wind speeds
9 to 14.9 rpm. are 4 and 25 m/s, respectively. Therefore, the rotor rotational
speed can range from 9 to 14.9 rpm.
2.3. Drive Train Kinematics
2.3. Drive Train Kinematics
The drive train kinematics must first be determined to identify the basic mechanical vibrations
inThe drive train kinematics
a rotor-dynamic system, whichmust first be
include determined
shaft imbalancetoandidentify the basicOutlines
gear meshing. mechanical vibrations
of the gearbox in
configuration system,
a rotor-dynamic for thewhichplanetary,
includeintermediate,
shaft imbalance and andhigh
gearspeed
meshing.stages are provided
Outlines in
of the gearbox
Figure 2. The
configuration rotor
for the shaft is the planetary
planetary, gearing
intermediate, andinput,
highwhile
speedthe output
stages areis provided
the low speed shaft. The
in Figure 2. The
rotor
rotor carries
shaft is thethe three planet
planetary gearsinput,
gearing that revolve around
while the the is
output sunthegear.
lowThe planet
speed gears
shaft. Thealso mesh
rotor withthe
carries
an outer ring gear. For the two parallel axis stages, the low speed shaft connects
three planet gears that revolve around the sun gear. The planet gears also mesh with an outer ring with the intermediate
shaft,
gear. Forwhich
the two in parallel
turn connects with the
axis stages, high
the lowspeed
speedshaft.
shaft connects with the intermediate shaft, which
The angular
in turn connects withrotation
the highfrequency of a shaft, f (Hz), is obtained based on the rotational speed,
speed shaft.
N The
(rpm), while using
angular Equation
rotation (1). of a shaft, f (Hz), is obtained based on the rotational speed,
frequency
N (rpm), while using Equation (1). 𝑓= . (1)
N
f = . (1)
The nomenclature used for the rotation frequencies
60 of the shafts and planets relative to the rotor
is The
given in Table 1. The
nomenclature particular
used for the kinematics of the studied
rotation frequencies gearbox
of the shafts can
andbe described
planets based
relative to on
thethe
rotor
number of gear teeth indicated in Table 2 and outlined in Figure 2.
is given in Table 1. The particular kinematics of the studied gearbox can be described based on the
number of gear teeth indicated in Table 2 and outlined in Figure 2.
Machines 2018, 6, 63 4 of 16
Machines 2018, 6, x FOR PEER REVIEW 4 of 16

(a) (b)
Figure2.2.(a)
Figure (a)Front
Frontview
view of
of the
the planetary
planetarygear;
gear;(b)
(b)Lateral
Lateralview ofof
view thethe
entire gearbox.
entire gearbox.

Table
Table 1. 1. Nomenclatureadopted
Nomenclature adopted to
to identify
identify the
the angular
angularrotation
rotationfrequencies of of
frequencies thethe
drive train
drive shafts.
train shafts.

Rotor RotorPlanet
Planet LowLow
Speed Intermediate
Speed High Speed High Speed
Intermediate
f0 fP f1 f2 f3
f0 fP f1 f2 f3

Table 2. Nomenclature and numbers of teeth for each gear.

Table 2. Nomenclature and numbers of teeth for each gear.
Sun Planet Ring Gear 1 Gear 2 Gear 3 Gear 4
Sun ZS
Planet ZP Ring
ZR ZGear
12 1 Z21 GearZ223 Z32 3
Gear Gear 4
18 34 87 70 16 84 19
ZS ZP ZR Z12 Z21 Z23 Z32
18 34 87 70 16 84 19
As a result of the gear process, typical vibrations at the gear mesh frequencies are generated in
the gearbox. The gear mesh frequencies of the planetary, intermediate, and high speed stages have
been
As named
a result gmf P, gmf
of the 12, process,
gear and gmftypical
23, respectively.
vibrationsThe main
at the rotation
gear frequencies are
mesh frequencies andgenerated
gear mesh
in the
frequencies are indicated in Table 3 as a function of the rotor frequency, f0. These frequencies can be
gearbox. The gear mesh frequencies of the planetary, intermediate, and high speed stages have been
calculated
named gmf P using12Equations 23
, gmf , and gmf (2)–(8).
, respectively. The main rotation frequencies and gear mesh frequencies
are indicated in Table 3 as a function of the rotor frequency, f 0 . These frequencies can be calculated
𝑓 =𝑓 . (2)
using Equations (2)–(8).
Zs + ZR
f 1 = f𝑍0 . (2)
𝑓 = (𝑓Z−
s 𝑓 ). (3)
𝑍
Zs
fP = ( f 1 − f 0 ). (3)
𝑓ZP= 𝑓. (4)
Z12
f2 = f1 . (4)
𝑓 =Z21 𝑓 . (5)
Z23
3 = = 𝑍f 2𝑓. .
f𝑔𝑚𝑓 (6) (5)
Z32
gm f P ==Z𝑍P f P𝑓. .
𝑔𝑚𝑓 (7) (6)
gm f 12 = Z12 f 1 . (7)
𝑔𝑚𝑓 =𝑍 𝑓. (8)
gm f 23 = Z23 f 2 . (8)
Table 3. Rotation and gear mesh frequencies as functions of the rotor frequency, f0.
Table 3. Rotation and gear
Rotation mesh frequencies
Frequency [Hz] as functions
Gear MeshofFrequency
the rotor frequency,
[Hz] f 0.
f0 fP f1 f2 f3 gmfP gmf12 gmf23
Rotation Frequency [Hz] Gear Mesh Frequency [Hz]
1 f0 2.56 f0 5.83 f0 25.52 f0 112.83 f0 86.97 f0 408.3 f0 2143.6 f0
f0 fP f1 f2 f3 gmfP gmf 12 gmf 23
1 f0 2.56 f 0 5.83 f 0 25.52 f 0 112.83 f 0 86.97 f 0 408.3 f0 2143.6 f 0
Machines 2018, 6, 63 5 of 16

Machines 2018, 6, x FOR PEER REVIEW 5 of 16

2.4. Specifications of the Vibration Monitoring System

2.4. Specifications of the Vibration Monitoring System
TheThewind
windturbines
turbinesarearebuilt
builtusing
using anan integrated system that
integrated system thatmonitors
monitorsthe themain
main operating
operating
parameters, including wind speed, power output, rotating velocity, blade pitch
parameters, including wind speed, power output, rotating velocity, blade pitch angle, and other angle, and other
factors.
factors. TheThe
windwind speedisismeasured
speed measuredwithwith ultra-sonic
ultra-sonic anemometers
anemometerslocated
locatedonon
thethe
roof of the
roof nacelle.
of the nacelle.
TheThe data
data forfor
allall
thethe turbinesininthe
turbines thewind
windfarm
farm are
are stored
stored in aa SCADA
SCADAsystem.
system.Measurements
Measurements areare
registered
registered every
every 1010 min.The
min. TheSCADA
SCADAdata datasets
sets are
are available
available for
foranalysis.
analysis.
A CM
A CM system
system basedononvibrations
based vibrationsisisinstalled
installed inin each
each wind
windturbine.
turbine.This
Thissystem
system conditions
conditionsandand
registers
registers thethe signals
signals fromfrom various
various accelerometers
accelerometers that
that areare mounted
mounted onon
thethe drive
drive train
train components.
components. Two
Two analogue
analogue inputs monitor
inputs monitor the outputthe power
output andpowertheand the generator
generator rotatingrotating
speed. speed. The system
The system is
is remotely
remotelyvia
controlled controlled
TCP-IP via TCP-IP (Transmission
(Transmission Control Protocol-Internet
Control Protocol-Internet Protocol),
Protocol), and theand the measured
measured data are
data are downloaded
downloaded at fixed intervals. at fixed intervals.
This
This studystudy focused
focused on seven
on seven industrial
industrial accelerometers
accelerometers mounted mounted
along thealong
drivethe drive
train train
components
components in the positions indicated in Figure 3. Sensors S1–3 are specifically designed for low
in the positions indicated in Figure 3. Sensors S1–3 are specifically designed for low speed applications.
speed applications. The remaining sensors are typical for rotating machinery applications. Two
The remaining sensors are typical for rotating machinery applications. Two sensors are mounted in
sensors are mounted in radial directions on each rotor bearing, three sensors are mounted in radial
radial directions on each rotor bearing, three sensors are mounted in radial and axial directions on the
and axial directions on the gearbox casing, and two sensors are mounted in radial directions on the
gearbox casing, and two sensors are mounted in radial directions on the generator. The positions of
generator. The positions of the sensors were specifically selected to monitor all shafts, bearings, and
thegears,
sensors
as were specifically
indicated in Tableselected
4. to monitor all shafts, bearings, and gears, as indicated in Table 4.

Figure3.3.Sensor
Figure Sensor locations
locations along
alongthe
thedrive
drivetrain.
train.

Table
Table 4. 4.Description
Descriptionof
ofthe
thesensor
sensor positions
positions and
anddetection
detectionobjectives.
objectives.

Sensor Sensor Location

Location Orientation
Orientation Detection Objective
Detection Objective
S1 Rotor Radial Front bearing
S1 Rotor Radial Front bearing
S2 Rotor Radial Rear bearing
S2 Rotor Radial Rear bearing
S3 Gearbox Vertical Planetary stage
S3 Gearbox Vertical Planetary stage
S4 Gearbox Vertical Low speed–intermediate stage
S4 Gearbox Vertical Low speed–intermediate stage
S5 Gearbox Axial Intermediate–high speed stage
S5 Gearbox Axial Intermediate–high speed stage
S6 Generator Radial Front side
S6 Generator Radial Front side
S7 S7 Generator
Generator Radial
Radial Rear side Rear side

2.5. Description of the Measurements

2.5. Description of the Measurements
The raw time signals from each sensor are digitized and processed using a digital signal
The raw (DSP)
processor time signals
withinfrom
the each sensor
online CM are digitized
system. Theand
DSPprocessed
calculatesusing
the afrequency
digital signal processor
spectrum.
Subsequently, the last measured time signal and the corresponding average frequency spectrum are
(DSP) within the online CM system. The DSP calculates the frequency spectrum. Subsequently,
thesaved in the datatime
last measured memory.
signal and the corresponding average frequency spectrum are saved in the
In addition, the CM system can calculate the frequency content of a vibration signal with user-
data memory.
defined broad band
In addition, the and
CMnarrow
systemband
can characteristic
calculate thevalues. The root
frequency mean square
content (RMS) value,
of a vibration ARMS
signal ,
with
is calculated from the frequency spectrum by adding the squares of the amplitudes
user-defined broad band and narrow band characteristic values. The root mean square (RMS) value, of N values over
a defined
ARMS frequency
, is calculated range
from the between
frequencyf1 and f2 and then
spectrum taking the
by adding the square
squaresroot of this
of the value, as indicated
amplitudes of N values
in Equation (9).
Machines 2018, 6, 63 6 of 16

over a defined frequency range between f 1 and f 2 and then taking the square root of this value, as
indicated in Equation (9). r
1
N ∑ f1 − f2
A RMS = A2 . (9)

The CM system processes different measurements depending on the input signals. In our case,
the baseband of the calculated frequency spectrum has been selected based on the major monitored
frequencies, which are dependent on the corresponding shaft rotation speeds, as indicated in Table 1.
Consequently, the adequate sampling frequency, fs , the cut-off frequency, fcutoff , the resulting frequency
resolution, ∆f, and the reference shaft frequency have been set for each measurement, as indicated in
Table 5.

Table 5. Measurements, sensors, sampling frequencies, cut-off frequencies, frequency resolutions, and
reference frequencies.

Measurement Sensor fs [Hz] fcutoff [Hz] ∆f [Hz] Reference Frequency

Rotor Bearing 1 S1 300 100 0.073 f0
Rotor Bearing 2 S2 300 100 0.073 f0
Gearbox 1 S3 600 200 0.147 f0
Gearbox 2 S4 3000 1000 0.732 f1
Gearbox 3 S5 6000 2000 1.465 f2
Generator 1 S6 6000 2000 1.465 f3
Generator 2 S7 6000 2000 1.465 f3
Gearbox 4 S5 30 10 0.007 f0

Measurements in the seven WTs were conducted for a continuous period of six months. Therefore,
more than 12,000 characteristic values and 150 spectra are available for each unit.

3. Data Validation
To guarantee that the investigated wind turbines were not affected by malfunctions during
the study period, the steady-state power output curves from the SCADA data set were compared
with those that were provided by the manufacturer following the procedure detailed in Escaler and
Mebarki [18]. The results of this comparison verified that the seven WTs were presumably operating
correctly and under optimal conditions.
Because the CM system only registers the power output and the speed of rotation at the moment
of the vibration measurement, the 10 min average wind speed SCADA data have been used to estimate
the corresponding wind speed at the instant of the measurement. The timestamps that are contained
in both databases have been used to interpolate the SCADA values.
The validity of the interpolated values of wind speed, as representatives of the instantaneous
values at the moment of the vibration measurement, was verified prior to their use. This verification
is necessary given the unsteady nature of the wind. Wind gusts and wind turbulence are among the
major causes of fluctuations in operating conditions and vibrations.
To identify and remove the vibration values that were not associated with steady wind conditions,
the power output was plotted as a function of the estimated instantaneous wind speeds and compared
with the upper and lower percentiles (5 to 95%) of the mean power curves, which were previously
calculated from the SCADA data set. Then, only the vibration measurements associated with the
output power levels plotted between the estimated percentiles were used for wind speeds from 4 to
12 m/s, meanwhile at wind speeds greater than 12 m/s, only those measurements with output power
values ranging from 1900 to 2100 kW were used, as described in Escaler and Toufik [18].
Machines 2018, 6, 63 7 of 16

4. Results
To identify the main frequency components of a vibration spectrum for a given sensor
measurement (Table 5), the spectrum frequency axis, f, has been divided by the corresponding
Machines 2018, 6, x FOR PEER REVIEW 7 of 16
instantaneous rotation frequency, fi , of the reference shaft (i = 0, 1, 2, or 3) at the time of the measurement.
From the spectra of all WTs with normalized frequency axes, the central amplitude value, or the 50th
measurement. From the spectra of all WTs with normalized frequency axes, the central amplitude
percentile, has been plotted in black combined with a grey area denoting the values that lie between
value, or the 50th percentile, has been plotted in black combined with a grey area denoting the values
thethat
25thlieand 75th percentiles. The median was used to compute average amplitude values, because
between the 25th and 75th percentiles. The median was used to compute average amplitude
it isvalues,
not appreciably
because it isinfluenced by outlier
not appreciably values
influenced and it is
by outlier generally
values and it more robustmore
is generally thanrobust
mean-based
than
estimates.
mean-based estimates. Thus, the 50th percentile spectra and the percentiles obtained for eachare
Thus, the 50th percentile spectra and the percentiles obtained for each measurement
measurement
plotted in Figures are4–8
plotted in Figures
and are 4–8 in
discussed and are
the discussedsections.
following in the following sections.

Figure
Figure 4. Median
4. Median spectrum(black
spectrum (blackline)
line) and
and amplitude
amplitude levels
levelsbetween
betweenthe
the25th
25thand 75th
and percentiles
75th percentiles
(grey region) of the vibrations measured on the front rotor bearing.
(grey region) of the vibrations measured on the front rotor bearing.

Figure
Figure 5. Median
5. Median spectrum(black
spectrum (blackline)
line) and
and amplitude
amplitude levels
levelsbetween
between25th
25thand 75th
and percentiles
75th (grey
percentiles (grey
region) of the vibrations measured on the rear rotor bearing.
region) of the vibrations measured on the rear rotor bearing.
Machines 2018, 6, 63 8 of 16
Machines 2018,
Machines 2018, 6,
6, xx FOR
FOR PEER
PEER REVIEW
REVIEW 88 of
of 16
16

Figure
Figure
Figure 6. Median
6. Median
6. Median spectrum(black
spectrum
spectrum (blackline)
(black line)and
line) and amplitude
and amplitude levels
amplitude levels between
levelsbetween 25th
between25th
25thand
and
and75th
75th percentiles
percentiles
75th (grey
(grey
percentiles (grey
region)
region)
region) of the
of the
of the vibrations measured
vibrationsmeasured
vibrations measuredon on the
onthe gearbox.
thegearbox.
gearbox.

Figure 7. Median spectrum (black line) and amplitude levels between 25th and 75th percentiles (grey
Figure 7. Median spectrum(black
(blackline)
line)and
and amplitude
amplitude levels
Figure 7. Median spectrum levelsbetween
between25th
25thand
and75th percentiles
75th (grey
percentiles (grey
region)
region) of the
of the vibrations measured
vibrationsmeasured
measuredon on the
onthe generator.
thegenerator.
generator.
region) of the vibrations
Machines 2018, 6, 63 9 of 16
Machines 2018, 6, x FOR PEER REVIEW 9 of 16

Figure
Figure 8. Median
8. Median spectrum(black
spectrum (blackline)
line) and
and amplitude
amplitude levels
levelsbetween
between25th and
25th and75th percentiles
75th (grey
percentiles (grey
region) of the low frequency vibrations measured in the axial direction on the gearbox.
region) of the low frequency vibrations measured in the axial direction on the gearbox.

4.1.4.1. Rotor
Rotor Vibration
Vibration SignatureAnalysis
Signature Analysis

The The vibratory signatures

vibratory signatures ofof thethe vibrations measured
vibrations on bothon
measured rotor bearings
both rotorare plotted in
bearings Figures
are plotted
4 and 5, reaching up to 600 times f0 . In both bearings, a frequency peak at 3f 0 ,
in Figures 4 and 5, reaching up to 600 times f 0 . In both bearings, a frequency peak at 3f 0 , whichwhich is called the blade
passingthe
is called frequency, fB, is clearly
blade passing observed.
frequency, fB ,This vibration
is clearly is due to This
observed. towervibration
dam excitation,
is duewhich occurs
to tower dam
excitation, which occurs once per revolution due to the short-term collapse of the aerodynamic the
once per revolution due to the short-term collapse of the aerodynamic forces at the blade when forces
blade approaches the tower.
at the blade when the blade approaches the tower.
In the front rotor bearing closer to the hub (Figure 4), the low frequency content is dominated
In the front rotor bearing closer to the hub (Figure 4), the low frequency content is dominated by f 0 ,
by f0, fB, and several harmonics of fB. In the high frequency range, f3 and 2f3 are also present,
fB , and several harmonics of fB . In the high frequency range, f 3 and 2f 3 are also present, corresponding
corresponding to the generator shaft. Finally, the gear mesh frequency of the planets, fP, and its
to the generator shaft. Finally, the gear mesh frequency of the planets, fP , and its harmonics appear
harmonics appear with sidebands at fB, as does the gear mesh of the low speed-intermediate
with gmf12. at fB , as does the gear mesh of the low speed-intermediate stage gmf 12 .
sidebands
stage
In In
thetherear
rearrotor
rotorbearing
bearing closer
closer to tothe
thegearbox
gearbox (Figure
(Figure 5), the
5), the vibration
vibration peakspeaks and amplitudes
and amplitudes are
aresimilar
similartotothose
thosethat
thatwerewereobserved
observedininthe thefront
frontbearing.
bearing.Nevertheless,
Nevertheless, the harmonic
the harmonic content content
of fof
B
fB
decreases, as only B2fB is present.
decreases, as only 2f is present.

4.2.4.2.
Gearbox Vibration
Gearbox VibrationSignature
SignatureAnalysis
Analysis
TheThe vibratory
vibratory signatures
signatures ofofthe
thevibrations
vibrationsthat thatwere
weremeasured
measured in in the
the gearbox
gearbox casing
casing are
are plotted
plotted in
in Figure 6. In the vertical position close to the planetary gear (top graph), the main peaks are
Figure 6. In the vertical position close to the planetary gear (top graph), the main peaks are associated
associated
with gmfP (upwith gmf
to its P (up
6th to its 6th harmonic),
harmonic), f 3 , 2f 3 , andf3gmf
, 2f3,12and
. Ingmf 12 . In general,
general, the frequency
the frequency is analogous
is analogous to those
to those in
measured measured
the rotorinbearings.
the rotor bearings.
With
With a basebandup
a baseband uptoto600600times
times f f11,, the
the highest
highest peakpeakininthe thevertical
verticalposition
positionclose to to
close thethe
intermediate shaft bearing (middle graph) corresponds to the
intermediate shaft bearing (middle graph) corresponds to the gear mesh frequency gmf 23 ofgear mesh frequency gmf 23 of thethe
intermediate-highspeed
intermediate-high speedstage.
stage. At
At lower
lower levels,
levels, peaks
peaks thatthatareareassociated
associatedwith withgmf 12 (up to its 6th
gmf 12 (up to its 6th
harmonic), gmf P, and f3 can also be found.
harmonic), gmfP , and f 3 can also be found.
With a baseband up to 440 times f2, the highest peaks in the axial position close to the generator
With a baseband up to 440 times f 2 , the highest peaks in the axial position close to the generator
shaft (bottom graph) are again gmf23, 2gmf23, and 3gmf23. The gmf12 and its harmonics up to 6gmf12 are
shaft (bottom graph) are again gmf 23 , 2gmf 23 , and 3gmf23 . The gmf 12 and its harmonics up to 6gmf 12
also measured.
are also measured.
4.3. Generator Vibration Signature Analysis
4.3. Generator Vibration Signature Analysis
The vibratory signatures of the vibrations measured on the generator supports are plotted in
The vibratory signatures of the vibrations measured on the generator supports are plotted in
Figure 7. In both the front and rear positions relative to the gearbox, the most remarkable peaks are
Figure 7. In both the front and rear positions relative to the gearbox, the most remarkable peaks are the
the high speed shaft frequency f3 and a harmonic component at 84 times f3 with sidebands at f3. The
high speed shaft
latter is not frequency
a gear f 3 and a harmonic
mesh frequency; component
therefore, it at 84 times
may be provoked f 3 with sidebands
by electromagnetic at f 3 . The
excitation. latter
Peaks
is not
canabegear mesh frequency;
observed around 84ftherefore, it may be provoked by electromagnetic excitation. Peaks can
3; however, these peaks cannot be attributed to a known kinematic
be condition
observed that is associated
around 84f 3 ; with the gearbox
however, these peaksor cannot
shaft. be attributed to a known kinematic condition
that is associated with the gearbox or shaft.
4.4. Tower/Nacelle Vibration Signature Analysis
4.4. Tower/Nacelle Vibration Signature Analysis
The vibratory signature of the low frequency vibrations measured on the gearbox casing close
to The
the vibratory
generator signature
in the axial direction
of the are plotted
low frequency in Figure
vibrations 8, reaching
measured up gearbox
on the to 15 times f0. close
casing This to
the generator in the axial direction are plotted in Figure 8, reaching up to 15 times f 0 . This measurement
Machines 2018, 6, 63 10 of 16

Machines 2018, 6, x FOR PEER REVIEW 10 of 16

quantifies the tower/nacelle vibrations, as suggested by Escaler and Mebarki [18]. The predominant
measurement
peak quantifies
is fB ; however, the tower/nacelle
individual vibrations,
values exhibit as suggested
significant by throughout
separation Escaler and Mebarki [18]. The
the spectrum. The
predominant peak is f B; however, individual values exhibit significant separation throughout the
other peaks correspond to f 0 , 2fB , and 3fB .
spectrum. The other peaks correspond to f0, 2fB, and 3fB.
5. Discussion
5. Discussion
5.1. Comparison of Mean Vibration Amplitudes and Deviations
5.1. Comparison of Mean Vibration Amplitudes and Deviations
The widths of amplitude values between the 25th and the 75th percentiles were significant for
most of The widths of amplitude
the frequency peaks thatvalues between the
were identified in25th and the 75th
the vibration percentiles
signatures were
of all thesignificant for
measurements.
most of the frequency peaks that were identified in the vibration signatures of all the measurements.
This result indicates significant spread or variability exists in the spectral values collected in this study.
This result indicates significant spread or variability exists in the spectral values collected in this
Therefore, dispersion was quantified for the most relevant peaks, while the mean spectrum and its
study. Therefore, dispersion was quantified for the most relevant peaks, while the mean spectrum
standard deviation were calculated from all the WT measurements. The corresponding results are
and its standard deviation were calculated from all the WT measurements. The corresponding results
plotted in Figures 9 and 10, with bars showing the average values and vertical lines showing the
are plotted in Figures 9 and 10, with bars showing the average values and vertical lines showing the
positive deviations.
positive deviations.

Figure
Figure 9. Mean
9. Mean vibration
vibration acceleration
acceleration RMSRMS and
and standarddeviation
standard deviationvalues
valuesfor
forthe
themain
mainfrequencies
frequencieson
the on
lowthe low speed
speed side ofside
theofdrive
the drive
train.train.

Similar
Similar vibration
vibration peaks
peaks and and
levelslevels are measured
are measured in bothinrotor
bothbearings,
rotor bearings,
althoughalthough more
more harmonics
thatharmonics that are
are associated associated
with the bladewith the blade
passing passingare
frequency frequency
presentarein present
the front inbearing
the frontclose
bearing close
to the hub,
to the hub, as shown in Figure 9. The generator frequency and the planetary gear
as shown in Figure 9. The generator frequency and the planetary gear frequency were also measuredfrequency were also
andmeasured and theirexhibit
their amplitudes amplitudes exhibitstandard
significant significant standard deviations.
deviations. The gearboxThe1 gearbox 1 measurement,
measurement, which was
which was located on the low speed side of the drive train, detected f 3 and gmfP , but at lower
located on the low speed side of the drive train, detected f 3 and gmfP , but at lower amplitudes than on
amplitudes than on the rotor bearings. This gearbox measurement detects up to the 6th harmonic of
the rotor bearings. This gearbox measurement detects up to the 6th harmonic of gmfP , as it has a larger
gmfP, as it has a larger frequency band than the other two rotor bearing measurements. In general, the
frequency band than the other two rotor bearing measurements. In general, the maximum vibration
maximum vibration levels are dominated by f0, fB, and the harmonics of fB in the rotor
levels are dominated by f 0 , fB , and the harmonics of fB in the rotor bearing vibrations.
bearing vibrations.
Machines 2018, 6, 63 11 of 16
Machines 2018, 6, x FOR PEER REVIEW 11 of 16

Figure
Figure 10. 10. Mean
Mean vibration
vibration accelerationroot
acceleration rootmean
mean square
square (RMS)
(RMS) and
andstandard
standarddeviation
deviationvalues forfor
values thethe
main frequencies on the high speed side of the drive train.
main frequencies on the high speed side of the drive train.

TheThe
twotwo measurements
measurements gearboxes
gearboxes 2 and
2 and 3, corresponding
3, corresponding to to sensors
sensors S4S4 and
and S5 S5 located
located onon
thethe
high
high speed side of the gearbox, detect the gear mesh frequencies of the intermediate and
speed side of the gearbox, detect the gear mesh frequencies of the intermediate and high speed stages,high speed
stages, in
as shown as Figure
shown 10.in The
Figure 10. The selected
baseband basebandforselected for the2 measurement
the gearbox gearbox 2 measurement (vertical
(vertical orientation)
orientation) encompasses gmf23 but not the associated harmonics. This measurement also detects gmfP
encompasses gmf 23 but not the associated harmonics. This measurement also detects gmfP and f 3 .
and f3. The larger frequency band of the gearbox 3 measurement (axial orientation) detects the
The larger frequency band of the gearbox 3 measurement (axial orientation) detects the harmonic
harmonic contents of gmf23 and gmf12. Note that the mean amplitudes of the peaks are similar in both
contents of gmf 23 and gmf 12 . Note that the mean amplitudes of the peaks are similar in both gearbox
gearbox measurements. Among all of the peaks, the highest amplitudes are observed for gmf23. For
measurements. Among all of the peaks, the highest amplitudes are observed for gmf 23 . For this
this meshing frequency, the standard deviation of the amplitudes is also significant.
meshingSignificant
frequency,standard
the standard deviation
deviation values of thealso
have amplitudes
been foundis also
for fBsignificant.
in the gearbox 4 measurement,
Significant standard deviation values have also been found
which monitors the tower vibrations, and for 84f3 in both generator for fBmeasurements,
in the gearboxwhich
4 measurement,
monitor
the electromagnetic vibrations.
which monitors the tower vibrations, and for 84f 3 in both generator measurements, which monitor the
electromagnetic vibrations.
5.2. Wind Speed Amplitude Dependency
5.2. Wind Speed Amplitude Dependency
To investigate the spread of individual amplitude levels observed in the majority of the
frequency peaks,the
To investigate thespread
calculated characteristic
of individual valueslevels
amplitude (basedobserved
on Equation
in the(9)) of the of
majority measurement
the frequency
basebands
peaks, (Table 5)characteristic
the calculated have been analyzed
values as a function
(based of wind (9))
on Equation speed. These
of the values represent
measurement the
basebands
broadband
(Table RMS analyzed
5) have been amplitudesas aobtained
function by adding
of wind all the
speed. individual
These frequencies
values represent thepresent
broadbandin the
RMS
spectrum.obtained
amplitudes In this case,
by data fromall
adding three
the of the seven frequencies
individual WTs is used, present
because they
in theutilize exactlyIn
spectrum. thethis
samecase,
gearbox model. The preliminary data validation that is discussed in Section 3 was
data from three of the seven WTs is used, because they utilize exactly the same gearbox model. The also applied.
The vibration
preliminary amplitudes
data validation that for one measurement
is discussed in Sectionin the rotor
3 was bearing,
also applied.one in the gearbox and one
inThe
the vibration
generator amplitudes
are plotted inforFigure 11. The amplitudes
one measurement in the are clearly
rotor dependent
bearing, on the
one in the operational
gearbox and one
conditions, as highlighted in all graphs. Moreover, it is confirmed that the data trend is dependent
in the generator are plotted in Figure 11. The amplitudes are clearly dependent on the operational
on the component being measured.
conditions, as highlighted in all graphs. Moreover, it is confirmed that the data trend is dependent on
For example, in the front rotor bearing, the data points exhibit a linear trend with a positive
the component being measured.
constant slope. In the high-speed side of the gearbox casing and in an axial direction, the data trend
For example, in the front rotor bearing, the data points exhibit a linear trend with a positive
is not linear. A rapid amplitude increase is observed at low wind speeds up to the rated wind speed,
constant slope. In the high-speed side of the gearbox casing and in an axial direction, the data trend is
peaking at approximately 7 m/s. Generally, the shape of the data resembles the output power curve
notoflinear. A rapid
the turbine. amplitude
From the ratedincrease
speed toisthe
observed
cut-out at low the
speed, wind speeds
data up toisthe
amplitude rated wind
relatively speed,
constant.
peaking at approximately 7 m/s. Generally, the shape of the data resembles the output power curve
Machines 2018, 6, 63 12 of 16

Machines 2018, 6, x FOR PEER REVIEW 12 of 16

of the turbine. From the rated speed to the cut-out speed, the data amplitude is relatively constant.
Finally, in the
Finally, rear
in the side
rear ofofthe
side thegenerator,
generator,thethe data
data trend is also
trend is also similar
similartotothe
theturbine
turbine power
power curve;
curve;
however, thethe
however, generator data
generator datahave
havea adifferent
differentslope
slopethan
than the
the gearbox data.
gearbox data.

Figure 11. Broad band RMS vibration amplitudes as a function of wind speed.
Figure 11. Broad band RMS vibration amplitudes as a function of wind speed.

In Figure
In Figure 12,12, it is
it is observedthat
observed thatthe
themedian
median and
and the
the percentiles
percentilesofofvibrations
vibrationsthat were
that weremeasured
measured
on the rear rotor bearing of two WTs are very similar. In fact, it is confirmed that the same happens
on the rear rotor bearing of two WTs are very similar. In fact, it is confirmed that the same happens for
for the rest of WTs and measurements. Consequently, it can be assumed that the large variability
the rest of WTs and measurements. Consequently, it can be assumed that the large variability found is
found is mainly due to the operating conditions. At low wind speeds, the spectra have low
mainly due to the operating conditions. At low wind speeds, the spectra have low amplitudes and
amplitudes and they are below the 25th percentile values. At high wind speeds, the spectra have high
they are below the 25th percentile values. At high wind speeds, the spectra have high amplitudes
amplitudes above the 75th percentile values. Therefore, high reference values do not indicate faults,
above
theythe 75th
just percentile
show values.
the vibration Therefore,
response high reference
to stronger loads onvalues do train
the drive not indicate faults, they just show
components.
the vibration response to stronger loads on the drive train components.
Machines 2018, 6, 63 13 of 16
Machines 2018, 6, x FOR PEER REVIEW 13 of 16

The
Theinformation
informationprovided
providedbyby
these plots
these highlights
plots highlightsthethe
difficulty of protecting
difficulty variable-speed
of protecting and
variable-speed
variable-pitch machines while using the CM system, because the amplitudes of the vibrations
and variable-pitch machines while using the CM system, because the amplitudes of the vibrations strongly
strongly
depend ondepend on the conditions.
the operation operation conditions. Nevertheless,
Nevertheless, this studythis study confirms
confirms the feasibility
the feasibility of
of obtaining
obtaining statistical information that represents the normal range of vibration levels for a set of
statistical information that represents the normal range of vibration levels for a set of similar WTs.
similar
With theseWTs. With limits,
vibration these vibration limits,and
a more precise a more precise
reliable and reliable
assessment of theassessment of the machine
machine vibration response
vibration
can response can be achieved.
be achieved.

Figure12.
Figure 12.Median
Medianspectrum
spectrum(black
(blackline)
line) and
and amplitude
amplitude levels
levels between
between25th
25thand
and75th
75thpercentiles
percentiles(grey
(grey
region) of the vibrations measured on the rear rotor bearing of two wind turbines (WTs).AAspectrum
region) of the vibrations measured on the rear rotor bearing of two wind turbines (WTs). spectrum
abovethe
above the75th
75thpercentile
percentilecorresponds
correspondsto tohigh
high wind
wind speeds,
speeds, and
and aa spectrum
spectrumbelow
belowthe
the25th
25thpercentile
percentile
correspondstotolow
corresponds lowwind
windspeeds.
speeds.

6.6.Fault
FaultDetection
DetectionCase
Case
AAfault
faultononthe
thedrive
drivetrain
trainand
and its
its symptoms
symptoms that
that is
is based
based on on the
the vibration
vibrationsignature
signatureofofthese
these
turbines
turbines is presented
is presented here.
here. TheThe
toothtooth
of gearof gear 2 in
2 in the the intermediate
intermediate shaft ofshaft
a windof turbine
a wind (seeturbine
Figure(see
2b)
Figure
was broken.2b) This
was can
broken.
be seenThis canphotograph
in the be seen in shown
the photograph
on Figure 13 shown
takenon Figure
during 13 takenThis
inspection. duringfault
inspection.
could This fault
be predicted andcould be predicted
repaired prior to a and repaired
machine prior to down
unexpected a machinetime unexpected
based on the down time
amplitude
based onof
evolution the amplitude
the evolution
corresponding gearofmesh
the corresponding
frequency, gmf gear meshitsfrequency,
12 , and first threegmf 12, and its first three
harmonics.
harmonics.
In Figure 14, the reference mean vibration acceleration RMS values of the gear mesh frequency
In Figure
peaks (Figure 10)14,
and thethe
reference mean vibration
corresponding acceleration
mean values measured RMSinvalues
the wind of the gear mesh
turbine frequency
with progressive
peaks (Figure 10) and the corresponding mean values measured in the
levels of gearbox damage are compared. The mean values presented in Figure 14 correspond to wind turbine with progressive
levels ofperiods
different gearboxofdamage
time alongare the
compared.
damageThe mean values
evolution. presented
Each value in Figure
has been 14 correspond
calculated from a settoof
different periods of time along the damage evolution. Each value has
around 300 measurements taken during an interval of four days. It must be noted that, during been calculated from a set of
those
around 300 measurements taken during an interval of
days, the wind turbine was operating normally without any restriction. four days. It must be noted that, during those
days, the wind turbine was operating normally without any restriction.
It can be seen that all the amplitudes of the gear mesh frequency peaks increased simultaneously
It can be seen that all the amplitudes of the gear mesh frequency peaks increased simultaneously
up to twice the reference levels when the gear damage started to develop. The peaks continued to
up to twice the reference levels when the gear damage started to develop. The peaks continued to
increase, as the damage developed, reaching their maximum amplitudes for advanced damage. Finally,
increase, as the damage developed, reaching their maximum amplitudes for advanced damage.
after repairing the fault, the amplitudes returned to their original levels.
Finally, after repairing the fault, the amplitudes returned to their original levels.
To observe the change of vibratory signature induced by the fault, one spectrum measured at full
To observe the change of vibratory signature induced by the fault, one spectrum measured at
load during a healthy period of time has been compared with one spectrum that was measured during
full load during a healthy period of time has been compared with one spectrum that was measured
the initial stages of damage, as shown in Figure 15. It can be clearly seen that all of the frequency peaks
during the initial stages of damage, as shown in Figure 15. It can be clearly seen that all of the
related to the gear meshing suffer a significant amplitude increase. Moreover, additional frequency
frequency peaks related to the gear meshing suffer a significant amplitude increase. Moreover,
peaks appearfrequency
additional which arepeakssideband
appear modulations.
which are sideband modulations.
Machines 2018, 6, 63 14 of 16

Machines 2018, 6, x FOR PEER REVIEW 14 of 16

Therefore, it is demonstrated that knowing a reference vibration signature with a mean value
and theTherefore, it is demonstrated
possible range that to
of variation due knowing a reference
operating vibration
conditions signature
is a valuable toolwith a mean value
to monitor changes
and the possible range of variation due to operating conditions is a valuable tool to
of the dynamic behavior of the drive train components induced by faults. To confirm the abnormal monitor changes
of the dynamic
behavior, behavior
the evolution withof time
the drive
of thetrain components
anomalous induced
levels must beby tracked.
faults. ToAs
confirm the abnormal
the damage develops,
behavior, the evolution with time of the anomalous levels must be tracked. As the damage develops,
these values will rise and a machine inspection should be carried out urgently before the failure occurs.
these values will rise and a machine inspection should be carried out urgently before the failure
Nevertheless, an additional study of the symptoms that are associated to other types of damages and
occurs. Nevertheless, an additional study of the symptoms that are associated to other types of
affecting other components is required to confirm the current findings and to evaluate more precisely
damages and affecting other components is required to confirm the current findings and to evaluate
their sensitivity.
more precisely their sensitivity.
more precisely their sensitivity.

Figure
Figure 13. Photograph
13. Photograph of 2 gear
of gear 2 at intermediate
at intermediate shaft with shaft withteeth
a broken a found
brokenout
teeth found
during out
inspection.
during inspection.
The variability of the reference vibration signature does not prevent detection of the initial stages
The variability
of the damage because, of as
thealready
reference vibration it
explained, signature
mainly does
takesnot prevent
into detection
account of the
the effect initial
of the stages
wind loads
of the damage because, as already explained, it mainly takes into account the effect
on the measured vibration levels. The median spectra of each healthy WT are within the 25th of the wind loadsand
75thonpercentiles
the measured
(greyvibration
area) oflevels. The median spectra
the corresponding of each
reference healthy
signature, asWT are within
shown the of
at the top 25th and 16.
Figure
75th percentiles (grey area) of the corresponding reference signature, as shown at the top of Figure
Meanwhile, the median spectrum of the damaged WT is clearly above the 75th percentile level, as
16. Meanwhile, the median spectrum of the damaged WT is clearly above the 75th percentile level,
shown at the bottom of Figure 16, which demonstrates the sensitivity of the presented reference values
as shown at the bottom of Figure 16, which demonstrates the sensitivity of the presented reference
to this particular gear damage.
values to this particular gear damage.

Figure
Figure 14.14. Comparisonbetween
Comparison betweenthe
the reference
reference mean
mean vibration
vibrationacceleration
accelerationRMS
RMSvalues of of
values thethe
gear
gear
mesh
mesh frequency
frequency and
and itsitsharmonics
harmonics(measurement
(measurement gearbox
gearbox 3)
3)and
andthe
thecorresponding
correspondingmean values
mean in ain a
values
particular
particular windwind turbine
turbine withprogressive
with progressivelevels
levels of
of gearbox
gearbox damage.
damage.
Machines 2018, 6, 63 15 of 16

Machines 2018,
Machines 6, x6,FOR
2018, PEER
x FOR REVIEW
PEER REVIEW 15 of1516of 16

Figure
Figure 15. 15. Comparison
Comparison between
between oneone spectrum in
spectrum in healthy
healthycondition and and
condition another one with
another oneinitial
with initial
gearbox damage corresponding to gearbox 3 measurement.
Figure 15. Comparison between one spectrum in healthy condition and another one with initial
gearbox damage corresponding to gearbox 3 measurement.
gearbox damage corresponding to gearbox 3 measurement.

Figure 16. Comparison of the reference median spectrum (black line) and amplitude levels between
25th and 75th percentiles (grey region) of the gearbox vibrations obtained from all the WTs with the
particular median spectra of each healthy WT (top graph) and with the median spectrum of the
damaged
Figure WT (bottomof
16. Comparison graph).
the reference median spectrum (black line) and amplitude levels between
Figure 16. Comparison of the reference median spectrum (black line) and amplitude levels between
25th and 75th percentiles (grey region) of the gearbox vibrations obtained from all the WTs with the
25th7.and
Conclusions
75th percentiles (grey region) of the gearbox vibrations obtained from all the WTs with
particular median spectra of each healthy WT (top graph) and with the median spectrum of the
the particular median spectra of each healthy WT (top graph) and with the median spectrum of the
The mean
damaged vibration
WT (bottom signature (baseline frequency spectrum) of a group of healthy WTs was
graph).
damaged WTusing
obtained (bottom graph).
accelerometers that were mounted on the rotor bearings, gearbox casing, and
generator supports. From these measurements, the frequency peaks that are associated with the
7. Conclusions
7. Conclusions
various excitations that originated at the WT components were identified. These frequencies are due
to The mean vibration signature (baseline frequency spectrum) of a group of healthy WTs was
mechanical, aerodynamic and(baseline
electromagnetic forces.spectrum)
These results
The mean
obtained
vibration signature frequency ofcan be usedofashealthy
a group referenceWTs was
spectrausing accelerometers
for assessing that were
normal machine mounted
conditions and on thealarm
to set rotor bearings,
levels. gearbox
If faults occur orcasing,
wear isand
obtained using supports.
generator accelerometers
From that were
these mounted on
measurements, thethe rotor bearings,
frequency peaks thatgearbox casing, and
arenormal
associated withgenerator
the
enhanced, new defect frequencies will appear and/or unexpected changes of the frequencies
various
supports. excitations
From
present these that originated
on the reference spectra ofatthe
measurements, the
theWT components
frequency
affected werebeidentified.
peaks
component will that are These frequencies
associated
observed. The with
recognition are
thedue
of these various
to mechanical,
excitations aerodynamic
that originated and
at the WT electromagnetic
components forces. These resultsThese
were identified. can befrequencies
used as reference
are due to
spectra aerodynamic
mechanical, for assessing normal machine conditions
and electromagnetic and These
forces. to set alarm
resultslevels.
can beIf faults
used occur or wearspectra
as reference is
enhanced,
for assessing new defect
normal frequencies
machine will appear
conditions and toand/or unexpected
set alarm levels.changes
If faultsofoccur
the normal
or wearfrequencies
is enhanced,
presentfrequencies
new defect on the reference
will spectra
appearofand/or
the affected component
unexpected will beof
changes observed.
the normal The recognition
frequenciesofpresent
these on
the reference spectra of the affected component will be observed. The recognition of these changes is
essential for the early detection of faults or threats. Moreover, a significant impact of the wind speed
on the vibration amplitudes has been observed; this impact also depends on the component being
Machines 2018, 6, 63 16 of 16

measured. Therefore, this behavior must also be taken into account when defining individual reference
values and predicting the severity of damage.

Author Contributions: X.E. has carried out the data analysis, prepared the figures and tables and written the
draft manuscript. T.M. has installed and configured the condition monitoring systems and collected the data.
Both X.E. and T.M. have discussed the results and drawn the conclusions.
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Yang, W.; Tavner, P.J.; Crabtree, C.J.; Feng, Y.; Qiu, Y. Wind Turbine Condition Monitoring: Technical and
Commercial Challenges. Wind Energy 2014, 17, 673–693. [CrossRef]
2. Hameed, Z.; Hong, Y.S.; Cho, Y.M.; Ahn, S.H.; Song, C.K. Condition monitoring and fault detection of wind
turbines and related algorithms: A review. Renew. Sustain. Energy Rev. 2009, 13, 1–39. [CrossRef]
3. Hameed, Z.; Ahn, S.H.; Cho, Y.M. Practical aspects of a condition monitoring system for a wind turbine
with emphasis on its design, system architecture, testing and installation. Renew. Energy 2010, 35, 879–894.
[CrossRef]
4. Gray, C.S.; Watson, S.J. Physics of failure approach to wind turbine condition based maintenance. Wind Energy
2010, 13, 395–405. [CrossRef]
5. Kusiak, A.; Li, W. The prediction and diagnosis of wind turbine faults. Renew. Energy 2011, 36, 16–23. [CrossRef]
6. Tian, Z.; Jin, T.; Wu, B.; Ding, F. Condition based maintenance optimization for wind power generation
systems under continuous monitoring. Renew. Energy 2011, 36, 1502–1509. [CrossRef]
7. Hau, E. Wind Turbines: Fundamentals, Technologies, Application, Economics, 2nd ed.; Springer: Berlin,
Germany, 2006.
8. García Márquez, F.P.; Tobias, A.M.; Pinar Pérez, J.M.; Papaelias, M. Condition monitoring of wind turbines:
Techniques and methods. Renew. Energy 2012, 46, 169–178. [CrossRef]
9. Zaher, A.; McArthur, S.D.J.; Infield, D.G. Online Wind Turbine Fault Detection through Automated SCADA
Data Analysis. Wind Energy 2009, 12, 574–593. [CrossRef]
10. Kusiak, A.; Zhang, Z. Analysis of Wind Turbine Vibrations Based on SCADA Data. ASME J. Sol. Energy Eng.
2010, 132, 1–12. [CrossRef]
11. Ghoshal, A.; Sundaresan, M.J.; Schulz, M.J.; Pai, P.F. Structural health monitoring techniques for wind turbine
blades. J. Wind Eng. Ind. Aerodyn. 2000, 85, 309–324. [CrossRef]
12. Yang, W.; Tavner, P.J.; Wilkinson, M.R. Condition monitoring and fault diagnosis of a wind turbine
synchronous generator drive train. IET Renew. Power Gener. 2009, 3, 1–11. [CrossRef]
13. Sheldon, J.S.; Watson, M.J.; Byington, C.S. Integrating oil health and vibration diagnostics for reliable wind
turbine health predictions. In Proceedings of the ASME 2011 Turbo Expo: Turbine Technical Conference and
Exposition, Vancouver, BC, Canada, 6–10 June 2011; pp. 939–948.
14. Tang, B.; Liu, W.T.; Song, T. Wind turbine fault diagnosis based on Morlet wavelet transformation and
Wigner-Ville distribution. Renew. Energy 2010, 35, 2862–2866. [CrossRef]
15. Liu, W.Y.; Zhang, W.H.; Han, J.G.; Wang, G.F. A new wind turbine fault diagnosis method based on the local
mean decomposition. Renew. Energy 2012, 48, 411–415. [CrossRef]
16. Abouhnik, A.; Albarbar, A. Wind turbine blades condition assessment based on vibration measurements
and the level of an empirically decomposed feature. Energy Convers. Manag. 2012, 64, 606–613. [CrossRef]
17. VDI. VDI 3834 Part 1. Measurement and Evaluation of the Mechanical Vibration of Wind Energy Turbines and Their
Components. Onshore Wind Energy Turbines with Gears; VDI: Düsseldorf, Germany, 2009.
18. Escaler, X.; Mebarki, T. Wind speed dependency of low-frequency vibration levels in full-scale wind turbines.
J. Sol. Energy Eng. 2015, 137, 1–5. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).