Condition Monitoring of Wind Generators
Lucian Mihet Popa1, Birgitte-Bak Jensen2, Ewen Ritchie2 and Ion Boldea1
Department of Electrical Machines and Drives, Politehnica University of Timisoara, 1900 Timisoara - Romania1
Institute of Energy Technology, Aalborg University, 9220 Aalborg-Denmark2
Phone: 40-256-204402, E-mail: mihet@k.ro, bbj@iet.auc.dk, aer@iet.auc.dk, boldea@lselinux.utt.ro
Abstract – This paper focuses on the experimental investigation
for incipient fault detection and fault detection methods existing
in the literature, suitably adapted for use in wind generator
systems using Doubly Fed Induction Generators (DFIGs). Three
main experiments (one for stator phase unbalance, one for rotor
phase unbalance and one for turn-to-turn faults) have been
performed to study the electrical behaviour of the DFIG. The
article aims to provide wind generators with further
documentation for an advanced condition monitoring system, in
order to avoid undesirable operating conditions and to detect and
diagnose incipient electrical faults.
I. INTRODUCTION
The reduction of operational and maintenance costs is
continuously required of modern wind generators. In
particular, with plans for extending wind farms at sea (making
the generators more inaccessible), it is required to increase
reliability and production time, simultaneously with an
increase of service interval. An important factor enabling this
is to provide wind generators with advanced condition
monitoring systems and monitoring the generator during
operation. This will help to avoid undesirable operating
conditions, and to detect incipient faults in the components to
detect sensor and actuator faults.
Many sources show that bearing faults and stator insulation
breakdown causes the majority of machine failures. For small
induction machines, the stator winding insulation degradation
is one of the major causes [1, 15].
The induction machine can operate under asymmetrical
stator and/or rotor-winding connections as described in [2, 3, 5
and 6]:
1. Inter-turn fault resulting in the opening or shorting of
one or more circuits of a stator phase winding;
2. Abnormal connection of the stator winding;
3. Inter-turn fault of a rotor phase winding and / or
brushes;
4. Broken rotor bars or cracked rotor end-rings (cage
rotor);
5. Static and/or dynamic air-gap irregularities;
6. A rub between the rotor and the stator which can result
in serious damage to the stator core and windings;
Faults in AC induction machines produce one or more of
the following symptoms:
• Unbalanced air-gap voltages and line currents
• Increased torque pulsation
0-7803-7883-0/03/$17.00 © 2003 IEEE
1839
• Decreased average torque
• Increased losses and reduction in efficiency
• Excessive heating
• Disturbances in the current/voltage/flux etc. waveform
Failure surveys have reported that percentage failure by
components in induction machines is typically [1, 6, 8]:
bearing related (40 %), stator related (38 %), rotor related (10
%), and others (12 %).
The objective of this paper is to develop and test methods of
condition monitoring, suitably adapted for implementation in
wind generator systems. This should contribute to increasing
the time in which the generator will produce energy, by
reducing the time spent in operating in undesirable load or
fault conditions, reduce the number of outages and provide an
improved background for planning service with relevant,
longer, service intervals.
II. DESCRIPTION OF THE EXPERIMENTAL SYSTEM
The experimental system is a model of a wind generator
system. The generator is a wound rotor induction generator
(DFIG), with slip rings, rated at 11 kW, as can be seen in
appendix, provided with a gearbox. The wind turbine rotor is
emulated by use of a drive system scaled for driving the
DFIG. The drive system is composed of a 15 kW squirrel cage
induction motor and a 22 kVA-frequency converter. Two
back-to-back PWM-VSI converters, with a dc link including a
dc capacitor filter are used to control the rotor current, active
and reactive power flow, as can be seen in Figure 1.
Freq. Conv. COND. MONITORING
SYSTEM
3~
IM Gear IG Back to back
Converter
Model of Wind Turbine Model of Generator System
Model of Wind Generator System
Fig. 1. Schematic diagram of the laboratory model of a wind generator
system.
 The most important component of the measurement system
is a complete condition monitoring system, which comprises
many subsystems like transducers, signal conditioning boxes
and data acquisition devices, as can be seen in Fig. 2.
Fig. 2. The block diagram of condition monitoring system.
A. The Condition Monitoring System
A complete condition monitoring system could comprise
many subsystems, each monitoring a particular part of the
wind generator. There will be a degree of overlapping between
the functions of the different subsystems.
This study includes the design and commissioning of a
measuring system comprising the AD Card - ICS 645, signal
conditioning boxes and transducers.
Transducers and sensors generate signals that must be
conditioned to provide the correct input to maximum the DAQ
device-ICS 645. The induction generator has been monitored
using measurements of stator and rotor currents, stator
voltages, rotor speed and temperature in the windings.
The Rogowsky Current Transducer-RGF 75 was selected to
measure stator and rotor currents because of its merits
compared to other transducers, such as: have a very wide
bandwidth, provide an isolated measurement at around ground
potential, can measure large current without saturating,
accuracy typically ± 1% and linearity ± 0.05% full scale /
0.1% actual reading.
The stator voltages were measured using a High Voltage
Differential Probe-P5200. The P5200 provides a safe means of
measuring circuits with floating potentials up to 1000 VRMS
from ground and up to 1300 V (DC + peak AC) differential.
The rotor speed was measured using a shaft incremental
encoder-Scancon 2R, which provide 9000 pulses/revolution.
The stator winding temperatures were measured by
inserting 12 thermocouples-Types J in various winding slots,
corresponding to each pole and phase of the induction
generator.
A1. The Signal Conditioning
Signal Conditioning provides the interface between the
signals/sensors and the measurement system. It improves the
performance and reliability of the measurement system with a
1840
variety of functions, including: signal amplification,
attenuation, isolation, filtering, multiplexing, linearization,
sensor conditioning, offsetting and noise reduction.
In our set up it was imperative to adapt the signals from
transducers as the DAQ – ICS 645 accepts the input signals
with peak amplitudes of ± 1.03 V with an input impedance of
500 Ohms. For example, thermocouples produce very low
voltage signals, which required amplification, filtering and
linearization. Voltage signals, acquired by high voltage
differential probes, required attenuation of the signal while the
voltage signals obtained from Rogowski coils required
filtering and offsetting of the signal. The output signals from
incremental encoder must be converted, using a frequency to
voltage converter, and then filtered and offsetted. All of these
preparation technologies are forms of Signal Conditioning.
A2. The Data Acquisition Device
Data acquisition devices are general purpose instruments
that are well suited for measuring voltage signals. The DAQ
ICS-645 is designed for high frequency, high precision, high-
density data acquisition and high-speed test and measurement
applications. It combines the ultimate in analog and digital
technologies to provide up to 16 channels and sample rates of
up to 5 MHz/channel. The AD Card – ICS 645 is able to
convert 16 channels simultaneously at up to 5 MHz sampling
frequency, and with 16 bit conversion. Based on a new
generation of ADC chip that combines Sigma-Delta and flash
converter technologies, the ICS-645 offers an exceptional
dynamic range over a wide bandwidth.
All recorded data were processed using the Matlab software
package to plot the currents and voltage spectra and then to
perform the FFT analysis. For each variable, 216 values were
recorded, with the sampling frequency of 2.5 MHz and with
the amount of data points of 32 768.
III. EXPERIMENTAL ARRANGEMENTS
Methods for the prediction of electrical behaviour in
deteriorating induction machines allow for an arbitrary
number of winding deterioration processes to be simulated on
the stator and / or the rotor. Forms of deterioration might
include high or low impedance between phases, between coils
in a single phase, or between a phase and ground [1].
Three experimental investigations have been done to study
the electrical behaviour of the induction machine: one stator
phase unbalance using a variable resistance and inductance in
series on one phase, one rotor phase unbalance using a
resistance of the same value as the rotor phase resistance
inserted in series on one rotor phase, and a turn-to-turn fault
using an inductance in parallel on one stator phase, as it is
depicted in Figures 3.
The stator of three-phase induction generator (DFIG) was
connected to the star connected power source, also in star
(wyes) connection.
 LV=(15 & 50) mH
The pictures below (Figs. 4, 5) show the rotor and stator
A A currents of the generator measured under a resistive unbalance
(Rv=1.2 Ω) in one rotor phase of the same value as the rotor
Switch on / off Switch on / off
phase resistance, corresponding to a stator power of 2 kW.
B B A very small difference exists in both stator and rotor
DFIG DFIG
currents and can only be due to the unbalance provoked.

C C

a1). a2).

LV=15 & 50 mH

b) c)

Fig. 3.a1) One stator phase inductive unbalance; a2) One stator phase
resistive unbalance;
b) One rotor phase resistive unbalance c) Turn-to-turn fault.

The procedure, which was used in creating an unbalance in Fig. 4. The rotor currents of DFIG acquired by TDS 540 under the resistive
one stator or rotor phase, consisted of increasing the unbalance of 1.2 Ω in one rotor phase at 2 kW. Scaling factor is 4 mV/A.
impedance of that phase using a variable resistance
(inductance) connected in series, as can be seen in Figs. 3a1,
3a2 and 3b. For instance, when a variable resistance of 10
Ohms was inserted between the grid and the stator phase
terminals on one phase, the stator impedance became 7 times
larger than in the balance case.
Placing an inductance and/or a resistance in parallel on one
phase, as shown in Fig. 3c), simulated deterioration of the
turn-to-turn insulation. The procedure consisted of decreasing
(weakening) the impedance of one stator phase by inserting a
variable resistance and/or inductance in parallel on that phase.
The early detection of various fault mechanisms could
prevent catastrophic failure from occurring and would also
allow for carefully planned repair actions [1, 2]. To achieve
this goal, on-line or off-line strategies that measure current
may be utilized.

IV. EXPERIMENTAL RESULTS

The measurements were acquired by a Tektronix

Oscilloscope-TDS 540 and by the DAQ device – ICS 645 via
Matlab software. The DFIG was operating, during the
measurements, at a rotor speed of 1475 rpm corresponding to
a stator active power of 2 kW. Fig. 5. The rotor speed and stator currents of DFIG acquired by ICS-645
under the resistive unbalance of 1.2 Ω in one rotor phase at 2kW.

1841
 Comparisons of Figures 6 and 7 show that the degree of
unbalance provoked in one stator phase, through introducing
the inductive unbalance of 50 mH (10 times more than stator
phase inductance), is reflected very clearly in one stator
current (Is1 in Fig. 6 on channel 2).
Fig.6. The stator currents acquired by TDS-540 under inductive unbalance-
Lv=50 mH at P=2 kW. The unbalance stator current is on channel 2. Scaling
factor is 4 [mV / A].
Fig.7. Speed and stator currents acquired by ICS-645 under inductive
unbalance-Lv=50 mH, at P=2 kW & n=1475 rpm. Is2: instantaneous
stator current of the unbalance phase.
1842
Fig. 8 shows that increasing significantly with a level of the
resistive unbalance in one stator phase (Rv=10 Ω) produce a
reduction of the stator current in that phase (ch. 2). It should
also be evident that the stator current decreasing on one phase
would provoke an increasing of others stator currents.
Fig. 8. Stator currents acquired by TDS-540 under one stator phase resistive
unbalance (Rv=10 Ω) at P=2kW and n=1475 rpm. The unbalance stator
current is configured on channel 2. Scaling factor is 4 [mV / A].
Fig. 9. Stator currents acquired by TDS-540 under one stator phase resistive
unbalance (Rv=1 Ω) at P=2kW and n=1475 rpm. The unbalance stator current
is configured on channel 2. Scaling factor is 4 [mV / A].
 A Tektronix Oscilloscope acquired the stator currents in
Figs. 6, 8 and 9. The measurement results show the efficiency
of line currents in identifying the presence of a resistive and an
inductive unbalance in one stator phase.
Figs. 8 and 9 show the stator currents of the induction
generator under two different levels of resistive unbalance
(Rv=10 Ω and Rv=1 Ω) in one stator phase. A comparison of
these Figs. point out that increasing significantly with a level
of the resistive unbalance in one stator phase (10 times larger
than phase resistance), produce a decreases of the stator
current of 14 times in that phase. Anyway, a clear difference
between balance and unbalance currents exist in both
situations very well.
In Fig. 7, the stator currents and rotor speed of DFIG were
acquired by the DAQ-ICS 645, for the same conditions as in
Fig. 6, with the sampling frequency of 2.5 MHz. Is1, Is2 and
Is3 denote instantaneous values of the line currents where Is2
represents the instantaneous stator current of unbalanced
phase.
Figs. 10-11 show the stator and rotor currents of the
generator that corresponds to stator power of 2 kW, under
turn-to-turn fault developed in one stator phase, and acquired
by a Tektronix Oscilloscope. Comparison of these Figs.
illustrate that the rotor currents (rms values) have almost the Fig. 11. The rotor currents acquired by TDS-540 under turn-to-turn fault in
same value while the imbalance created into the stator currents one stator phase (Lv=50 mH in parallel - Fig. 3c) at P=2 kW and n=1475 rpm.
is more clear. Hence, the stator current could be a good Scaling factor is 4 [mV / A].
indicator in detection of this type of fault.
Time-domain analysis using characteristic values to
determine changes by trend setting and for extracting the
amplitude information from current signals has been used in
this section as first evaluation tool. The acquisition of data by
Oscilloscope was done to comparison with AD Card
acquisition and to verify the accuracy and sensitivity of signal
conditioning box.

V. CURRENT SIGNATURE ANALYSIS TO DETECT

INDUCTION GENERATOR FAULTS

Machine current signature analysis (MCSA) is a non-

Fig. 10. The stator currents acquired by TDS-540 under turn-to-turn fault in
one stator phase (Lv=50 mH in parallel - Fig. 3c) at P=2 kW and n=1475 rpm.
Scaling factor is 4 [mV / A].
invasive, online or offline monitoring technique for the
diagnosis of problems in induction machines, such as turn-to-
turn fault [6, 7, 15], broken rotor bars [5, 6, 9, 11, 13], static or
/ and dynamic eccentricity [2, 8, 12, 13, 14].
Due to its powerful technique merits in diagnosis and
detection of faults the MCSA monitoring technique was
chosen as a fault detection method.
The stator and rotor currents monitoring system that has
been developed consists of three main sub-systems. These
include: signal conditioning, data acquisition and data
analysis, Fig. 12. Data acquisition and data analysis sections
are accessed by Matlab software package run from a PC. The
controlling software allows the operator of the system to store
the data in data files for future processing.

1843
 Transducer
Signal ADC
Conditioning PC
ICS-645

Data Acquisition Data Analysis

MATLAB

Fig. 12. Single channel acquisition system.

Figure 12 shows a diagram of a single channel of hardware

required to obtain the indicators for analysis purposes.
The data will first be processed to synthesise suitable
indicators. The indicators will then be resolved into spectra,
showing the frequency content of the indicators such as stator
and rotor currents. For Condition Monitoring, Fourier analysis
has been employed using the coefficients of the current 125 Hz

spectrum or the power spectrum as an indication of how the

harmonic content of a signal varies. If the variation of the
harmonic content can be related to specific faults then it may
be useful as an indicator.

Fig. 13. Spectrum of stator currents under turn- to-turn fault in one stator
phase at PG=2kW. The data were recorded and processed by Matlab software
A. MCSA to Diagnose Stator Turn-to-Turn Faults package and acquired via ICS-645.

The objective of this method is to identify current

components in the stator winding that are only a function of
shorted turns and are not due to any other problem or
mechanical drive characteristic. The following equation gives
the components in the air-gap flux waveform that are a
function of shorted turns [3, 5, 6, 7 and 15]:
n 
f st = f 1  (1 − s ) ± k  ………………………….(1)
 p 
fst = stator frequency components that are a function of
shorted turns, f1=supply frequency, n=1,2,3…
k=1,3,5,…p=pole-pairs, s=slip
The diagnosis of shorted turns via MCSA is based on
detecting the frequency components given by equation (1) in
that these rotating flux waves can induce corresponding
current components in the stator winding.
Figures 13 and 14 give the line current spectra for all stator
phases under turn-to-turn fault (Fig 13) and with no stator
fault (Fig. 14). The obvious change, in Figure 13, is that
completely new current components exist around 125 and 375
Hz and can only be due to the shorted turn, as per theory with
k=1, n=3 and k=1, n=13 (equation 1).
In addition the Figs. 13 and 14 make clear that the
difference between balanced operation and under turn-to-turn
fault cases follows the phase angle of stator current - Is1. For
instance the phase angle is changed around 125 Hz where a Fig. 14. FFT for a healthy machine of DFIG at PG=2kW, s=-0.016, VS=390 V,
IS=4.7 A. The data were recorded and processed by Matlab software package
new faulty component exist as well. and acquired via ICS-645.

1844
B. MCSA to Diagnose Stator Winding Unbalance
a)
175 Hz
175 Hz 425 Hz
175 Hz 425 Hz
b)
Fig 15. Stator currents spectrum under resistive (a) and inductive unbalance
(b) in one stator phase at PG=2 kW. The data were recorded and processed by
Matlab software and acquired by ICS-645.
Figs. 15 show a comparison between the stator currents
spectrum under one stator phase resistive unbalance (R=10 Ω)
and one stator phase inductive unbalance (L=50 mH). The
inductive unbalance creates the occurrence of new
components in the stator current spectrum at 175 Hz (k=1 and
n=5) and at 425 Hz (k=1 and n=15) while the resistive
unbalance provokes the occurrence of new components in the
stator current spectrum at 125 Hz (k=1 and n=3), 325 Hz (k=1
and n=11) and 375 Hz (k=1 and n=13) as well.
1845
C. MCSA to Diagnose Rotor Winding Unbalance
a)
125 Hz 375 Hz
375 Hz
125 Hz 375 Hz
b)
Fig. 16. Frequency Spectrum of the stator currents (a) and rotor currents (b)
under rotor unbalance in one rotor phase at PG=2kW. The data were recorded
and processed by Matlab software and acquired by ICS-645.
Figures 16 show the stator and rotor line currents spectrum
under rotor unbalance in one rotor phase. A clear difference in
the spectrum of the stator currents appear at 75 Hz, as can be
seen in Fig. 16a. Another new component appears at 375 Hz.
The rotor current spectrum offers new faulty components at
375 Hz (k=1 and n=13) and at 125 Hz (k=1 and n=3) as well
and sometimes at 325 Hz (k=1 and n=11), as can be seen in
Fig. 16b).
It can be concluded that the rotor line current spectrum,
under rotor unbalance in one rotor phase, offers more
information’s about occurrence of this fault than the stator line
current spectrum. Anyway, both of them may be used as an
indicator in detection of incipient rotor windings unbalance.
While, under the resistive and inductive unbalance in one
stator phase and under turn-to-turn fault in one stator phase,
the stator line current spectrum offers more information’s.
V. CONCLUSIONS
The objective of this paper was to develop different type of
faults and to test detection methods of condition monitoring,
previously reported in the literature, and suitable adapted for
implementation in wind generator systems. The results
presented have shown that the objective was achieved. An
induction machine condition monitoring system has also been
developed to test these methods. The software, which controls
both the acquisition and analysis of the signals, is written in
Matlab and has been developed as a sub-part of the monitoring
system. The article has shown that the detection of these faults
is also possible by time and frequency domain analysis. The
frequency spectrum of the stator and rotor line currents was
found to give the best results.
Placing an inductance and / or a resistance in series on one
phase, between the grid and stator phase terminals on one
phase, an unbalance in one stator phase was simulated.
Inserting a resistance in series on one rotor phase we also
simulated an unbalance in one rotor phase. Placing an
inductance in parallel on one phase, the deterioration of the
turn-to-turn insulation in one stator phase of the induction
generator was simulated. In this way we simulated a turn-to-
turn fault in one stator phase without destroying the machine
or building an additional construction such as placing taps on
the wires of the stator windings and then shorting these taps.
The experimental results show the efficiency of line stator
and rotor currents monitoring in identifying the presence of an
unbalance in one stator and rotor phase and confirm the
presence of harmonics as described by equation (1).
The experimental results have clearly demonstrated that
MCSA can diagnose turn-to-turn faults. MCSA can also
diagnose other problems in induction generators such as
inductive and resistive unbalance in one stator and rotor phase.
REFERENCES
[1] A.H.Bonnet and G.C.Soukup, “Cause and analysis of stator and rotor
failures in three-phase squirrel-cage induction motors”, IEEE Trans.
Ind. Applicat., vol. 28, pp. 921-937, July/August 1992.
[2] P.Vas, ”Parameter Estimation, Condition Monitoring, and Diagnosis
of Electrical Machines”, Clarendon Press, Oxford, 1993.
1846
[3] S.Nandi and H.A.Toliyat, “Fault Diagnosis of Electrical Machines – A
Review”, IEEE Industry Applications Conference, 1999. Thirty-
Fourth IAS Annual Meeting. Vol. 1, 1999, pp. 197-204.
[4] H.A.Toliyat, T.A.Lipo, “Transient Analysis of Cage Induction
Machines under Stator, Rotor Bar and End Ring Faults”, IEEE Trans.
On Energy Conv., Vol. 9, No. 4, June 1995.
[5] G.B.Kliman, W.J.Premerlani, R.A.Koegl and D.Hoeweler, “A new
approach to online fault detection in ac motors”, IEEE-IAS Annual
Meeting Conference, pp. 687-693, San Diego, CA, 1996.
[6] W.T.Thomson and M.Fenger, “Current Signature Analysis to Detect
Induction Motor Faults”, IEEE Industry Applications Magazine, pp.
26-34, July/August 2001.
[7] William T.Thomson, “On-line MCSA to diagnose shorted turns in low
voltage stator windings of 3-phase induction motors prior to failure”,
Electric Machines and Drives Conference, 2000. IEMDC 2001. IEEE
International, 2001, pp. 891-898.
[8] El Hachemi Benbouzid, M., “A review of induction motors signature
analysis as a medium for faults detection”, Industrial Electronics,
IEEE Transactions on, Volume: 47 Issue: 5, Oct. 2000, Page(s): 984
–993.
[9] W.T.Thomson, M.Fenger, “Industrial application of current signature
analysis to diagnose fault in 3-phase squirrel cage induction motors”,
Pulp and Paper Industry Technical Conference. Conference Record
of 2000 Annual Meeting, pp. 205-211.
[10] W.T.Thomson, “On-line current monitoring to detect electrical and
mechanical faults in three-phase induction motor drives”, Proc. IEE
and IMECHE (London), Int. Conf. Proc on Life Management of
Power Plants, Heriot-Watt University, Edinburgh, 12-14 December
1994, pp. 66-73.
[11] F. Filippetti, G.Franceschini, G.Gentile, S.Meo, A.Ometto,
N.Rotondale, C.Tassoni, “Current pattern analysis to detect induction
machine non rotational anomalies”, ICEM 98, International
Conference on Electrical Machines, September 2-4, 1998, Istambul-
Turkey, Vol. 1, pp. 448-453.
[12] S. Nandi, Hamid A. Toliyat and Alexander G. Parlos, “Performance
Analysis of A Single Phase Motor Under Eccentric Condition”, IEEE
Industry Applications Society, Annual Meeting, New Orleans,
Louisiana, October 5-9, 1997, pp. 174-181.
[13] Randy R. Schoen and Thomas G. Habetler, “Effects of Time-
Varying Loads on Rotor Fault Detection in Induction Machines”,
IEEE Transactions on Industry Applications, vol. 31, no. 4,
July/August 1995, pp. 900-906.
[14] J.R. Cameron, W.T. Thomson and A.B. Dow, “Vibration and current
monitoring for detecting airgap eccentricity in large induction
motors”, IEE Proceedings, Vol. 133, Pt.B, No. 3, May 1986, pp. 155-
163.
[15] J. Penman, H.G.Sedding, B.A.Lloyd, W.T.Fink, “Detection and
location of interturn short circuits in the stator windings of operating
motors”, IEEE Trans. Energy Conv., Vol. 9, no. 4, Dec. 1994,pp.
652-658.
APPENDIX:
TABLE. I.
NAMEPLATE DATA FOR LEROY SOMER 11 kW WOUND
ROTOR SLIP RING GENERATOR.
LEROY SOMER, MOT~FLSB 180 M4 B3, No. 14618900HG01, kg:
220
IP55 IK 1 cl. F 40 °C S1
V Hz 1/min KW cosϕ A
690Y 50 1460 11 0.81 IS=13
VR=1700 IR= 6.1
DE 6310 15cm3 11000 H
50/60 Hz
NDE 6310 15cm3 11000 H
50/60 Hz