Condition Monitoring and Fault Detection of Electric Drives 153

x8

Condition Monitoring and Fault

Detection of Electric Drives
Christian Kral1 and Thomas G. Habetler2
1Austrian Institute of Technology, 2Georgia Institute of Technology
1Austria, 2USA

1. Introduction
An electric drive consists of an electric machine, which converts electrical power to
mechanical power, power electronics to operate the machine and a unit to control the
motion of the drive. These are the components of the drive. Parts of each of these
components could fail and give rise to specific failure scenarios. The drive types
investigated in this chapter are limited to asynchronous induction machine and permanent
magnet synchronous machines, since these are the most common machine types in modern
electric drive applications. Faults of power electronics are not discussed since most failures
lead to the outage of the drive as the power electronics usually show no symptoms before
failure.
The task of identifying and classifying drive failures from certain measured quantities is
called fault detection. Under some conditions, fault detection may require certain safety
protection actions. Example: A turn to turn short circuit in the stator winding of the machine
is one example for a safety critical issue. If the short remains for a certain time, parts of the
winding will be destroyed. This in turn could cause winding failures that lead to a larger
short circuit current which may result in the failure and outage of the entire drive. In this
sense, a safety critical issue is a time critical issue. If the failure cannot be detected within a
certain time, the drive will be damaged and fails. It is thus highly demanded to accurately
detect safety critical faults and to protect the drive (and the application) in this case.

1.1 Classification of Methods

Condition monitoring of electric drives is motivated by different intentions. According to
these intentions methods can be classified:
 The operating conditions of the drive are stored in a specific format. From the
reported state a maintenance action or a warning for the operator can be triggered.
Such methods are condition monitoring methods.
 Some state quantities of the drive can be monitored to be fed back to the operating
strategy of the drive. Therefore, these classes of methods may be called condition
based control methods. An estimated operating temperature of, e.g., the winding of
the stator or rotor of the machine can be used to control the machine such way that
a certain temperature limit of the windings is not exceeded.

www.intechopen.com
154 Fault Detection

 A fault indicator can be used to trigger a protection switch to disconnect the drive
from the power supply (Farag et al., 1996). In this case the purpose of the condition
monitoring method is to protect the drive and the method is called a protection
method.
Condition monitoring and fault detection of electric drives are very important tasks in order
to maintain the reliability and safety of the drive. Additionally unexpected failures and
expensive repair and downtime costs can be avoided or limited this way.
From a technical point of view condition monitoring and fault detection methods can be
classified as passive (non-invasive) and active methods.
 Passive methods utilize measured quantities and derive certain fault or condition
indicators. These methods are usually applied to mains supplied machines and
have to no way of influencing the actual operating condition of the drive.
 Active methods do need any kind of actuator to control the condition of the drive.
Such an actuator may be either the inverter of the drive or an additional power
electronics device, connected in series to the feeder cables of the machine.
Another classification refers to physical domain for where the measured signals are
acquired from.
 Electrical monitoring evaluates only electrical signals such as the currents, voltages,
and – in some cases – the mechanical angle between the rotor and stator by means
of encoder. For the detection of faults with electrical origin, usually, it is
advantageous to evaluate electrical signals.
 Magnetic monitoring assesses the signatures determined from measurement coils.
Such coils may either be embedded into the stator slots of the machine or are
located externally as Rogowki coils (Henao et al., 2003).
 Vibrational monitoring is based on the evaluation of vibrational acceleration or
velocity. Mechanical imbalances and faults are difficult to detect by electrical
quantities. It is thus more reasonable to elaborate vibrational signals for the
detection of faults with mechanical origin.
 For thermal monitoring the desired temperature can either be measured directly or
estimated from state observers or models (Gao et al., 2008a; Kral et al., 2004a).
 Acoustic monitoring is based on the condition specific sound emissions of a drive
(Gaylard et al., 1995). The quality of acoustic monitoring, however, is very much
dependent on the background noise of the environment, the machine is operated
at. It is therefore very difficult to implement such methods in a real industrial
environment.
 For mixed physical domain monitoring methods signals from different domains are
evaluate and processed.
In this chapter only electrical monitoring methods are investigated and discussed.

1.2 Structure of Monitoring Methods

Monitoring methods are usually based on the processing blocks illustrated in Fig. 1. In the
first processing block the signals are measured. For currents and voltages usually current
and voltage transducers are used. The rotor speed and the rotor angle, respectively, are
either retrieved by an encoder or determined from the slotting harmonics of the currents in
the frequency or time domain (Hurst & Habetler 1996; Kral et al., 2006). Alternative

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 155

monitoring methods may measure other quantities such as temperatures, the vibrations of
the housing, acoustic emissions or the flux in internal or external coils.

Fig. 1. Block diagram of methods for the detection of machine conditions and faults

Signal conditioning includes some initial processing of the measured quantities. Such
processing includes the calculation of the phase voltages and the currents from the
measured quantities, depending on whether the connection of the machine is either wye or
delta. Additionally, for some methods the space phasors

vs 
2

v1  e j 2 / 3 v 2  e  j 2 / 3 v 3 
 
(1)
3
i s  i1  e j 2 / 3 i 2  e  j 2 / 3 i3
2
(2)
3

are determined from the phase voltages v1 , v2 , v3 and phase currents i1 , i2 , i3 .

The evaluation method incorporates knowledge on the computation of internal machine
quantities or fault signatures from the measured signals. An internal quantity could be, e.g.,
the stator or rotor temperature of the machine. An example of a fault signature is the
harmonic component of the current that indicates an electrical asymmetry of the squirrel
cage of an induction machine. In this sense, physical or empirical models or practical
findings are applied to indicate the condition of the machine.
If a sound knowledge based on the physical context of certain conditions exists, the
assessment of the condition is straight forward. Example: The stator temperature of a
machine is estimated by means of condition monitoring technique. The insulation of the
stator winding is specified for a specific insulation class and maximum operating
temperature. In this context it is clear that if the maximum operating temperature is
exceeded, the machine may be damaged. If the stator temperature is integrated into the
control of the load of the machine, the set values for the load can be modified such way, that
maximum operating temperature is not exceeded. In another application one wants to
predict how long the machine can be operated if it is operated above the maximum
temperature. Even if the manufacturer of insulation materials provide life cycle curves
which assess the total life time as a function of the operating temperature, it may still be
difficult to forecast the remaining life time in an application. From this example the reader
can suspect that it is very difficult the give a precise answer to the question: How long is it

www.intechopen.com
156 Fault Detection

possible to operate the machine without failing, considering the given conditions or fault
condition. The final question in this context is: when does the operator have to schedule
maintenance for the machine. In the literature mostly condition monitoring and fault
detection methods are investigated without having the focus on fault assessment (Kral et al.,
2003). Only little effort has been spent on the prediction of the remaining life time of
machines so far – because the subject is so complex and time and cost intense.

2. Faults in Permanent Magnet Synchronous Machines

The high power density and high efficiency of the permanent magnet (PM) machine has led
to the use of this machine in applications in which the high reliability is a key feature, such
as aerospace/aircraft actuators, automotive auxiliaries and traction, and other industrial and
stand-by power generation applications. The switched reluctance motor drives are often
used for these applications because of their inherent fault tolerance (Mecrow et al., 1996).
While this is true, the switched reluctance machine has a lower power density compared to
the PM machine, and therefore is often undesirable. If PM drives are to be considered for
these applications, they have to be designed to be very reliable. Redundancy and
conservative designs have been used for improving the reliability of these drives against the
variety of faults that can occur (Bianchi et al., 1996). As an alternative to these expensive
solutions, considerable diagnostic strategies and control schemes have been devised to
ensure a fault tolerant drive.
Permanent magnet machines can be divided into two types: permanent magnet
synchronous machines (PMSMs) (which have sinusoidal induced stator voltages), and
brushless DC machines (BLDCMs) (which have trapezoidal induced stator voltages). PM
machines have been very desirable since their torque density and efficiency is higher than
any other machine type. However, like all machines, they can fail. Moreover, due to the
presence of rotating permanent magnets, damage to the machine can progress even if the
stator is disconnected from the line. Fault diagnosis, condition monitoring, and fault
tolerant operation of PM machines is very important, if not critical.

2.1 Detecting Faults in Permanent Magnets

Rotor magnet defects can be detected by observing the motor current for the same
frequencies as given by equation (7) below (Rajagopalan et al., 2004). They can also be
detected by estimating the strength of the permanent magnet as demonstrated for PMSMs in
(Le Roux et al., 2003). Estimating the instantaneous or RMS back-EMF gives a measure of
the flux linkages resulting from the rotor excitation, making it possible to estimate the
magnet strength. There are some examples in the literature of estimating the back-EMF, for
example in (Wang et al., 2001).
One method for detecting magnet problems is by estimating the d-axis magnet flux linking
the stator windings in the synchronous reference frame (Le Roux et al., 2003). This method
is derived in the rotor reference frame. The instantaneous value of the d-axis permanent
magnet flux linking the stator windings, d ( pm ) , is estimated by, (averaging the current over
r

an integer number of fundamental periods),

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 157

vqsr  rs iqsr
ˆdr ( pm )   Lrds idsr
r
(3)

In eqn. (3) all quantities are instantaneous values and idsr is the d-axis stator current, Lrds is
r
the d-axis stator inductance, vqs is the q-axis stator voltage, rs is the stator resistance, and
 r is the rotor speed. The average value of dr ( pm ) is a direct measure of the magnet
strength. In many applications, the rotor angle might not be available to determine the
currents and voltages in the rotor rotating reference frame. For these reasons, the estimation
is best done in the synchronous reference frame, by averaging ˆd ( pm ) , which is given by,
e

vqse  rs iqse
ˆde( pm )   Leds idse .
e
(4)

Fig. 2. Estimated magnet strength for different rotor faults at different operating conditions
(SL: 640 rpm, 3 Nm; SH: 640 rpm, 6 Nm; FL: 1280 rpm, 4.4 Nm, FH: 1280 rpm, 8.7 Nm)

This method to estimate the magnet strength with (4) was implemented on the
measurement data of a normal motor case and all the rotor fault cases. The results of this
estimation on these motors operating at different speed and loads are shown in Fig. 2.
For BLDC machines, magnet fault can be found by estimating the mean value of the torque
constant, Kt (Rajagopalan et al., 2004). For brushless dc machines (BLDC), this is simply the
mean value of the added back-EMFs of the two conducting phases in any particular 60-
degree rotor position region.
The method averages the supply voltage (V) and twice the DC-link current (Idc) multiplied
by the stator resistance (rs) in every 60-degree region of rotor position during steady-state
operation. However, this estimation neglects the variations in stator inductances. The
estimation works better when the back-EMF (E) is used directly, but this is not available on a

www.intechopen.com
158 Fault Detection

physical machine. However, the estimated magnet strength is independent of other faults
such as the dynamic and static eccentricities, implying that this estimation could still be
used to detect a decrease in the magnet strength. Thus, the estimation of the magnet
strength is given by,

V  2rs I dc
Kˆ  
E
r r
. (5)

2.2 Detection of Rotor Eccentricity

Load unbalance, misalignment, improper mounting, and a bent rotor shaft can all cause
rotor eccentricities. These eccentricities can be divided into static and dynamic eccentricities
(Fig. 3). Static eccentricity is when the rotor is shifted from its normal position at the center
of the stator and it rotates in that position. With dynamic eccentricity, the rotor is also
shifted from the normal position, but now rotates around the center of the stator. These
rotor faults cause problems such as vibration and noise due to unbalanced magnetic pull
(UMP). It also causes dynamic problems by adding to torque pulsations.

Fig. 3. (a) centric rotor, (b) rotor with static eccentricity, (c) rotor with dynamic eccentricity

Past research in the detection of rotor eccentricities in induction motors has shown that these
faults affect certain frequency components in the stator current which can be monitored for
use in diagnosing a fault (see section 3). These frequency components (Cameron et al., 1986)
and are given by equation (20) in section 3.4 below. Setting k, c, and o in (20) to the
appropriate integers yields the frequencies that have to be monitored to detect dynamic
eccentricity. Setting c = 0, yields the principle slot harmonics, which are the frequencies that
have to be monitored to detect static eccentricity. However (20) is of little practical use as
the knowledge of stator slots is not available and BLDC machines do not have rotor slots.
Dynamic eccentricity in induction motors causes current components at frequencies of,
(Dorrell et al., 1997),

f de  f e 
2mf e
. (6)
P

This equation is the same as equation (18) below, except with slip equal to zero. In (6), fde is
the dynamic eccentricity frequency, fe is the fundamental frequency, and m is any integer. In
the presence of dynamic and static eccentricity, current components at the rotating
frequency sidebands of fde can be monitored. This equation can be adapted for use in
diagnosis of permanent magnet machines too (Le Roux et al., 2002). In the case of BLDC
motor drives, there is no rotor slip. Also, there are no rotor bars or rotor windings and

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 159

therefore there are no rotor slots. This means that R = 0 and s = 0 in (6). Thus, the only
frequencies that can be used to detect dynamic eccentricity are those given in (6) and integer
multiples of the supply frequency harmonics, thus m.fe. The only frequencies that are
influenced by static eccentricity are the integer multiples of the supply frequency harmonics
(setting R = s = c = 0).

Motor current spectrum - Normal M otor

0.1
F undam ental (fe)
P hase current (A )

0.08

0.06 0.67fe 1.33fe

0.04

0.02

0
40 50 60 70 80 90 100 110 120 130 140

Motor current spectrum - D ynam ic E ccentricity

0.1
F undam ental (fe)
P hase current (A )

0.08
0.67fe
0.06

0.04
1.33fe
0.02

0
40 50 60 70 80 90 100 110 120 130 140
F requency (HZ)

Fig. 4. Comparison of good motor current spectrum to a motor with a dynamic eccentricity

The harmonic spectrum of a good 6-pole PMSM and the same machine with a dynamic
eccentricity are shown in Fig. 4. Note the presence of the sidebands of the fundamental at
the rotating frequency.

2.3 Detection of Stator Faults

The different types of stator turn faults are depicted in Fig. 5. The first stage is small shorts
between turns in the same winding. Among the five failure modes, turn-to-turn faults
(stator turn fault) have been considered the most challenging one since the other types of
failures are usually the consequence of turn faults. Furthermore, turn faults are very
difficult to detect at their initial stages.
Conventional turn fault detection schemes merely monitor the negative sequence
component of line currents (or the effective negative sequence impedance) and rely on
mathematical models for symmetrical induction machines to account for the effect of
unbalanced supply voltages on the negative sequence current (Kliman, 1996). However,
neglecting inherent asymmetries can lead to misdetection, with catastrophic consequences.
The issue of inherent asymmetries can be addressed by using a neural network-based
approach (Tallam et al., 2000).
While this technique has been shown to give very good results for line-connected induction
machines, it is often not suitable to PM synchronous machine applications since the current
controller in the drive attempt to regulate the current to track the reference value. Since the

www.intechopen.com
160 Fault Detection

current is regulated, the effect of the asymmetry from the fault is now reflected in the motor
voltage. For this reason, voltage-based turn fault detection methods have been proposed,
but require additional voltage sensors and cables.

Open Circuit

Turn-to-Turn

Coil-to-Coil

Line-to-Line

Line-to-ground

Fig. 5. Possible failure modes in wye-connected stator windings

By modeling a machine with a turn fault, it can be concluded that a bolted turn fault reduces
the positive sequence components of the machine impedances and back-emf voltages, while
increasing the negative sequence and coupling terms in the impedance matrix at the same
time. The positive sequence current slightly increases under a stator turn fault condition in
a mains-fed application where the power supply is a fixed voltage source (Lee et al., 2002).
In a Current Controlled Voltage Sourced Inverter (CCVSI)-driven application, the inverter
controls the line currents so as to follow their references by introducing negative sequence
voltage and reducing positive sequence voltage under a stator turn fault condition (Lee et
al., 2007). Since the inverter output voltages are produced according to the voltage
references that are generated through the current controllers, the variations in the machine
parameters will be reflected into the voltage references. This implies that for a given
rotating speed and current references (or alternatively torque reference), the presence of a
stator turn fault results in a reduced positive sequence component and an increased
negative sequence component of the voltage references as compared to a machine without a
turn fault. Thus, it can be concluded that the differences in positive and negative sequence
components of the voltage references, for a given torque reference and rotating speed, under
a stator turn fault and fault-free conditions can indicate the occurrence of a stator turn fault.
The voltage references in the rotating and stationary reference frame until fault and no-fault
conditions are shown in Fig. 6.

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 161

vqds e
vqds _ nom

e
vqds e
1
_ pos
vqds

2e
_ neg

Time
• Blue solid line : Stator voltage under fault-free condition
• Red dashed line : Positive sequence voltage with stator turn faults
• Red long-dashed line: Negative sequence voltage with stator turn faults

Fig. 6. Voltage references under fault-free and turn fault conditions

2.4 Fault Tolerant Operation

Turn faults are particularly problematic in Interior Permanent Magnet Synchronous
Machine (IPMSM) drives in safety-critical applications. This is due to the fact that the
rotating magnet can often not be stopped when a fault occurs, and therefore the fault
current is allowed to flow until catastrophic or dangerous thermal damage is done to the
machine. Therefore, PM machines not only require a reliable turn fault detection method,
but also imperatively require a proper remedial action that can maintain the drive’s
uninterrupted operation.
The most desirable characteristic of a remedial action is to maintain the drive’s
uninterrupted operation without any degradation in the performance characteristics of the
drive in the presence of a stator turn fault. Unfortunately, this is very difficult to achieve,
and only redundancy-based approaches can solve this difficulty. But these approaches can
be justified in specific applications. In transit applications such as traction drives, an
uninterrupted operation during a short period of time, even with a limp operation, can
prevent injury or death.
A simple stator turn fault-tolerant strategy for IPMSM drives that does not require any
hardware modification to the standard drive configuration has been proposed (Lee et al.,
2006). This strategy does not result in the complete loss of availability of the drive.
Generally, the asymmetry in the stator voltages resulting from a stator turn fault has only a
small effect on the overall stator voltage. Therefore the amplitude of the faulty phase
e
voltage is almost the same as that of the complex stator voltage vector ( vs ) in the
synchronous rotating reference frame. Consequently, current in the faulted winding is
given by,
v~se

 
if 
~
. (7)
 rs  j e Lls   ( L1  3L2 )
Rf ~


www.intechopen.com
162 Fault Detection

Where Rf is the resistance of shorted section of the winding,  is the fraction of the winding
that is shorted, Lls is the leakage inductance, and L1 is the average self inductance of the
winding, and L2 is the first harmonic magnitude of the self inductance. This equation
implies that an appropriate selection of q- and d-axis current combination for a given
operating condition can reduce the stator voltage significantly; consequently, a significant
reduction in i f is achievable while maintaining the given operating condition.

2.5 Fault Detection under Non-Stationary Conditions (Rajagopalan et al., 2005)

Operating conditions of an electric motor usually change rapidly over time. Diagnostics of
motor faults in such conditions is a challenging problem due to the need for application of
sophisticated signal processing techniques that can process non-stationary signals. While
some research has been reported in the detection of faults in induction motors operating
under very slowly varying speed and load conditions (Yazici et al., 1999 and Kim et al.,
2002), no research has been reported in the diagnosis of faults in motors operating under
rapidly varying operating conditions.
The classical technique for characterizing the time evolution of non-stationary signals is the
short-time Fourier transform (STFT), a linear time frequency representation (TFR) analysis
technique. Though simple and rugged, the STFT still assumes that the non-stationary signal
is slowly changing in the chosen time window. This and the choice of window length
impair the use of STFT at low frequencies or when the signal is changing very fast
dynamically. Beginning with the Wigner-Ville distribution a plethora of quadratic TFR
techniques have been introduced in an attempt to improve the time-frequency detail over
that achievable with the STFT (Marple et al., 1998). They have been extensively used in the
area of mechanical engineering for detection of gear faults.
The use of quadratic TFRs has been presented as a solution for the diagnostics of rotor faults
in electric motors operating under non-stationary load and speed conditions. Although the
method could be applied to any motor, its application is limited in this discussion to BLDC
motors. Four time-frequency representations have been considered – Wigner-Ville
distribution (WVD) (Cohen, 1989), Choi-Williams Distribution (CWD) (Choi et al., 1989),
and the Zhao-Atlas Marks Distribution (ZAM) (Zaho et al., 1990). The use of time-frequency
distributions has proven suitable for detection of rotor faults in electric motors operating
under continuous non-stationary conditions. However, the need to provide high frequency
resolution along with good cross-term suppression leads to complicated kernels requiring
large amounts of processing power.
Some of the commonly used TFRs have been implemented on a DSP platform to study their
computational loads. It is observed that quadratic TFRs such as WVD and CWD are
computationally more intensive than linear TFRs such as the spectrogram. The quadratic
TFRs however provide much better frequency resolution and localization of energy with the
ZAM distribution exhibiting the best performance. The CWD in particular is a good trade-
off between the excellent frequency resolution of the WVD and the high cross-term
suppression of the ZAM distribution. These distributions also do not depend on the size
and type of the window as in the spectrogram. The increased computational load is the
price paid if a better frequency resolution and good localization of energy is needed. In
spite of the increased complexity involved, the computation time of a CWD is still in the
order of a few tens of micro-seconds and hence is amenable to implementation in real-time.
This computational time can be further decreased by paralleling several micro-programmed

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 163

systems and using more optimized software routines. The results of Fig. 7 show that the
fault frequency magnitudes can be clearly seen even in the case of a time-varying
fundamental frequency for the machine.

CWD of filtered BLDC stator current (dynamic ecc., 5 Hz sine speed ref)
0.2
max
0.18

0.16
4/3 fundamental
0.14 rotor fault frequency
Frequency [kHz]

0.12

0.1

0.08

0.06

0.04

0.02 2/3 fundamental

rotor fault frequency min
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time [ms]
Fig. 7. CWD of a simulated BLDCM rotor fault.

3. Faults in Induction Machines

Many fault statistics have been published for induction machines, e.g., (Bell et al., 1985). In
these statistics the failure causes are classified. In most statistics the majority of failures are
due to bearing and stator faults. Since these statistics mostly refer to industrial applications
the statistics may look different for, e.g., railway traction drives. Additionally, any statistic is
inherently incomplete, since it very much depends on the boundary conditions and the
integrity of the determined data. However, induction machine conditions are usually
classified as:
 Stator faults are electrical faults in the stator winding of the machine. Stator faults
may be turn to turn faults, interturn faults, phase to phase faults, and phase to
ground faults.
 Rotor faults refer to electrical faults of the rotor of an induction machines. Since
most machines have a squirrel cage rotor, these faults are also called cage faults.
 The determination of the stator and rotor temperature is of particular interest for the
conditional monitoring and control of the machine. The temperature does not
directly indicate a fault, but may be an indicator of disadvantageous operating
conditions or severe stress to the machine.
 Eccentricity faults are caused by mechanical imbalances, misalignments, bent shafts,
or non-uniform air gaps. These eccentricities may be caused by the machine, the
mechanical coupling or the mechanical load or any combinations of these. To a
certain extent eccentricity faults can be detected by means of electrical
measurements over the reaction of magnetic field air gap on the stator currents.

www.intechopen.com
164 Fault Detection

 Bearing faults are due the failure of one of the bearings of the machine. The causes
of failures are manifold and detection of the origin is very difficult. Particularly, the
detection of bearing faults by means of electrical signals is a real challenge (Stack et
al., 2006).

3.1 Detection of Stator Faults

In the case of a fully symmetric stator winding the stator magneto motive force (MMF) can
be synthesized as a Fourier series, i.e., a fundamental and higher harmonic waves. A
symmetric winding gives rise to a constant magnitude of the fundamental MMF wave. A
fault of the stator winding leads to asymmetry of the stator magneto motive force (MMF)
caused by the stator winding. This asymmetry causes the magnitude of the stator MMF not
be constant any more. In this case the fundamental wave can be decomposed into a forward
and backward traveling wave with constant magnitudes. The forward traveling wave is
represented by the positive sequence component of the stator current i sp and the additional
backward traveling wave due to the stator fault is reflected by the negative sequence
component of the stator current i sn . The sequence components are computed from the
phase voltages and currents, however.
A negative sequence current could also be caused by a negative sequence voltage which is
due to a supply asymmetry. For the reliable detection of stator turn and winding faults it is
important not to confuse these faults with supply voltage imbalances. Nevertheless, the
measured negative sequence current and voltage component are used to identify stator turn
and winding faults (Tallam et al., 2007; Lee et al., 2003). Additionally, inherent winding
asymmetries have to be taken into account for stator fault detection techniques.
The negative sequence voltage, v sn is coupled to the positive and negative sequence current
through impedances,

v sn  Z snp i sp  Z snn i sn . (8)

In order to compensate inherent winding asymmetries the impedances Z snp and Z snn have
to be identified. These impedances are (slightly) dependent on the operating conditions of
the machine. The identification of the impedances may therefore be performed by means of
neural networks or functional approximations (Tallam et al., 2000). This way supply voltage
imbalances and inherent winding asymmetries are properly taken into account. For less
sensitive methods, the impedances are assumed to be constant quantities (Kral et al., 2007).
After a learning stage, where impedances are identified, the fault detection method can be
applied in the regular operation of the machine.
For the identification of the impedances Z snp and Z snn a negative sequence voltage is
required, since a symmetric voltage supply does not give rise to negative sequence currents
in case of a symmetric machine. The negative sequence voltage can be generated by means
of a single phase ohmic resistor which leads to asymmetric line to line voltages. If it is not
possible to introduce a negative sequence voltage, the impedances cannot be indentified –
except for inherent voltage asymmetries caused by asymmetric loads.
From the measured positive and negative sequence current, the computed negative
sequence voltage (8) can be compared with the measured negative sequence voltage.

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 165

Alternatively the computed negative sequence current according to (8), incorporating the
measured positive sequence current and the measured negative sequence voltage, can be
compared with the measured negative sequence current. Any deviation of the computed
from the measured value then indicates a stator fault.
In closed loop inverters the identification of the negative sequence component could be
contained in the machine voltages and currents (Tallam et al., 2003). It is thus important to
take both, the negative sequence voltage and current for the fault detection into account.
Nevertheless, the artificially generation of negative sequence currents for the determination
of the impedances applied in (8) is much easier since the reference currents can be
superimposed with a negative sequence reference quantity.
An alternative method for the detection of stator faults in inverter fed drives is based on the
statistical evaluation of the switching behavior of the current controller which is caused by
an electrical asymmetry of the stator winding (Wolbank et al., 2003). The advantage of this
method is that it easy to implement on an existing dive, since statistical evaluation has to be
implemented by means of additional software routines.

(a) (b)
Fig. 8. The squirrel cage rotor of an induction machine consists of (a) rotor bars and end
rings and (b) sheet iron

3.2 Detecting Faults in the Squirrel Cage

The squirrel cage of induction machine consists of R rotor bars and end rings on both ends
which are placed in the sheet iron of the rotor as depicted in Fig. 8. The bars and end rings
of small machines are usually die casted of aluminum or copper. For larger machines, the
copper bars are insulated and manually fitted into the sheet iron. In this case the bars and
end rings are either welded or hard-soldered.
In the symmetric cage of a squirrel cage induction machine the root mean square (RMS)
values of the bar currents are equal – except for inherent asymmetries. Therefore, also end
ring currents on both sides are equal. In case of a faulty rotor bar, which is physically
reflected in an increase of the respective bar resistance, the current flow of this bar is
diminished.

www.intechopen.com
166 Fault Detection

Fig. 9. RMS values of rotor bar currents for the case of one broken rotor bar (index 1); results
refer to an 18.5 kW induction machine with 40 rotor bars

In Fig. 9 the RMS rotor bar currents of a squirrel cage with one broken rotor bar are
depicted. Since bar number 1 is broken, the respective current is zero. Therefore, the bars
adjacent to the fault location have to carry higher RMS bar currents than the rest of bars. The
increased currents lead to an increased thermal stress of the respective bars. Due to this
phenomenon the adjacent bars could also fail. This in turn leads to the spreading out of the
fault from an initial fault location. Over time, the fault condition of the rotor gets worse and
worse until the machine fails.
In a real machine a faulty bar leads to an additional phenomenon. Since the rotor bars are
usually not insulated from the sheet iron, interbar currents occur. The original bar current is
thus diverted to the iron parts of the rotor which in turn damages the sheets.
Apart from broken rotor bars, broken end ring or broken junctions of the bars and end ring
segments may give also rise to an electrical asymmetry of the rotor. The entirety of these
faults is also called electrical rotor asymmetries. These faults cause a distortion of the rotor
MMF which can be interpreted as a forward and a backward traveling rotor MMF wave –
with respect to the rotor reference frame. The backward traveling rotor MMF wave induces
voltages in the stator winding at the so called lower side band frequency

r
f lsb  (1  2 s ) f e , (9)

where f e is the fundamental stator supply frequency and s represents slip. The superscript
r indicates that the frequency refers to a rotor fault. The interaction of this harmonic
component with the flux causes a double slip frequency

f tr  2 sf e (10)

torque pulsation. Due to the finite inertia of the drive, rotor speed pulsations occur at the
same frequency. The speed pulsations then induce additional side band harmonics in the
stator currents, i.e.,

[ m ]  (1  2ms ) f e ,
r
f lsb (11)

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 167

[ m ]  (1  2ms ) f e .
r
f usb (12)

where m is a positive integer order number. The index usb abbreviates the term upper side
band, however. The entirety of current harmonics is also reflected in low frequency torque
and instantaneous power pulsations at

f tr[ m ]  2msf e . (13)

A typical stator current Fourier spectrum for steady state operation of an induction machine
is depicted in Fig. 10. The equidistant displaced lower and upper side band currents are
clearly arising in this figure.

Fig. 10. Fourier spectrum of a per unit stator current during steady state operation; results
refer to a 18.5 kW induction machine with one broken rotor bar out of 40 bars

A rotor fault can only be detected when the rotor currents lead to sufficient reactions on the
air gap field. The magnitudes of the side band currents of stator current are proportional to
load torque in steady state. The magnitudes of side band harmonics with low ordinal
numbers m are larger than the magnitudes with higher ordinal numbers. When the
mechanical load of the machine is relieved, slip decreases and the magnitudes of the side
band currents decrease as well. In this case the side band harmonics merge with the
fundamental according to (11) and (12). Therefore, a rotor fault cannot be detected when the
machine is mechanically unloaded in steady state. For the reliable detection of rotor faults a
minimum load torque of approximately 30% of the nominal torque is required.
The magnitudes of the lower and upper side band harmonics are also influenced by the total
inertia of the drive. A large inertia leads to a dominant magnitude of the lower side band
harmonic component (9) in the stator current, whereas a low inertia gives rise to a decreased
lower side band and an increased higher side band component (Kral et al., 2008a).
Most rotor fault detection methods are based on the evaluation of one or more stator current
signatures and are thus called current signature analysis (CSA) methods. For steady state
operating conditions a conventional Fourier analysis can be used to identify electrical rotor
asymmetries. For such applications, it is important to determine the magnitudes of the
respective current harmonics with high precision. Since low frequency torque fluctuations

www.intechopen.com
168 Fault Detection

can also induce current sidebands as depicted in Fig. 10, it is important to exactly track the
fault specific frequency components so that they do not get confused with load specific
harmonics (Kral et al., 2005). From the magnitudes of the side band harmonics the extent of
the fault can be estimated according to formulas presented in the literature, e.g., (Thomson
& Ranking 1987; Bellini et al., 2000; Culbert & Rhodes 2007).
During transient operating conditions, load torque and thus slip are not constant. A
conventional Fourier analysis is therefore not applicable in this case. The problem of time
varying load torque can either be solved by means of compensating the modeled load
torque (Schoen et al., 1995) or by mathematical transformations and algorithms (Wu et al.,
2005). Other approaches use a short time Fourier analysis or multiple signal classification
(MUSIC) algorithms to overcome the problem of identification of the non-stationary fault
signatures (Benbouzid & Kliman 2003). The fault detection under transient operating
conditions can also be performed by means of wavelet transforms (Antonino-Daviu et al.,
2006).

Fig. 11. Scheme of the Vienna Monitoring Method: V = voltage model, I = current model; F =
low pass filter for determining the (average) load torque from the torque of the voltage
model, C = spatial data clustering technique, D = discrete Fourier analysis

Apart from these methods, electrical rotor asymmetries can be detected by means of the
Vienna Monitoring Method (VMM). A scheme of this method is depicted in Fig. 11. This
method is a model based approach which compares the torque signals T v and T i of a
voltage and current model of the machine, resptively; the superscript indicates the model
reference. Both these mathematical models are the models of a symmetrical machine. The
voltage model is based on the stator voltage equation in the stator reference frame, whereas
the current model relies on the rotor voltage in the rotor reference frame. The input
quantities of the voltage model are the stator voltage and current space phasor (1) and (2).
Since the current model is utilized in the rotor reference frame the stator current space
phasor has to be transformed by means of the mechanical angle  m – which is the electrical
angle of the rotor with respect to the stator.
Applied to a fully symmetrical machine each model calculates the same torque and thus the
torque difference

T  T v  T i (14)

is zero. In case of a rotor fault, both models will calculate a double slip harmonic torque
oscillation – as well as higher harmonics which are, however, not taken into account in the
VMM. The double slip harmonic torque oscillations computed by the voltage and current
model show different magnitudes and phase angles, since both models have a different
model structure and different input signals. It turned out that the magnitude of the torque

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 169

difference is directly proportional to the fault extent and the load torque (Wieser et al., 1999).
If the torque difference is divided by the estimated load torque – which is determined from,
e.g., the voltage model – the relative torque difference is obtained,

T
t  . (15)
Tload

The magnitude of the double slip frequency component of t is independent of the load
torque and thus a measure for the electrical asymmetry of the rotor. Nevertheless, the
frequency of the relative torque difference is depending on slip. In order to eliminate the
time dependency of the relative torque difference, a spatial data clustering technique is
applied. This technique investigates the relative torque difference versus the angle of the
rotor flux space phasor,   ,which is determined by the current model in Fig. 11. Since the
rotor flux with respect to the rotor reference frame rotates with slip frequency, the double
slip frequency oscillation of the relative torque difference is mapped into a second harmonic
component with respect to the rotor circumference. A clustering technique is used to
average the wave forms of the obtained second harmonic signal. For this purpose, that rotor
circumference is subdivided into n data segments. The data value of each segment is then
derived by applying a recursive averaging algorithm. The data values represent an averaged
and discretized wave of the second harmonic torque difference. After a certain measurement
period a discrete Fourier analysis is applied to the data values. The magnitude of the second
harmonic then represents the averaged magnitude of the double slip frequency component
of the relative torque difference – which in turn serves as fault indicator for the VMM (Kral
et al., 2008a).
A great advantage of the VMM is that it reliably detects rotor faults under almost any
operation and boundary conditions. The VMM works
 independent of the load torque of the machine,
 under stationary and transient load conditions,
 under varying load and speed conditions,
 independent of the inertia of the drive,
 for mains supplied and inverter fed machines, independent of the applied control
technique and structure.

3.3 Monitoring of Stator and Rotor Temperature

The stator temperature of a machine can be measured by means of sensors which are
embedded in the winding, the winding heads or the sheet iron of the stator. This is a state of
the art measurement technology. Measuring the rotor temperatures is much more difficult
since the rotor is spinning and measuring lines can thus not be used. Even if there exist
radio frequency (RF) transmission sensors such technology is almost impossible to be
implemented in an industrial environment, because the machine to be monitored has to be
dismounted and modified – which is highly undesired.
In some applications it may be undesired to even measure the stator temperatures. In this
case it is then required to estimate the temperatures of the stator and rotor, respectively.
Neither stator nor rotor temperature are precise terms, since temperature is a local quantity.
Both the stator and the rotor have specific temperature distributions, depending on the

www.intechopen.com
170 Fault Detection

actual operating conditions. It is for the symmetrical machine (and supply), however, useful
to assume homogenous or averaged temperatures, for the conductors in the stator slots,
each winding head as well as each rotor bar and both end rings. By means of a lumped
parameter thermal model of the machine, these temperatures can be determined (Kral et al.,
2008b) – even online. In order to parameterize such a model all geometric details of the
stator, rotor, housing and cooling concept have to be known and modeled. A complex
lumped parameter model is thus not applicable in practice if the only little is known about
the machine.
For practical applications simplified lumped thermal parameter models can be used (Gao et
al., 2008b). These models usually employ only a couple thermal time constants. The thermal
time constants have to be determined during an identification test or are computed or
measured in advance before the machine is put into operation. A lumped parameter thermal
model can also be combined with any other resistance estimation method to a hybrid model
(Kral et al., 2004a).
In practice, only the average temperature of the stator winding can be estimated. The
average stator winding temperature increase is determined by estimating the stator
resistance, considering the temperature dependence of the conductor according to,

R  Rref (1   ref (T  Tref )) , (16)

where Rref is the resistance at temperature Tref , and  ref is the linear temperature
coefficient at the reference temperature. In this equation T is the actual temperature and R
is the respective actual resistance. One class of methods determines the average stator
winding temperature by means of a direct current (DC) bias. Since a DC current does not
cause any voltage drop across the stator inductance, the DC voltage represents the voltage
drop across the stator resistance only. For symmetric machines it can be assumed that the
winding temperatures are equal in all three phases. Therefore, the DC bias is injected only in
one phase. For mains fed induction machines the bias could be applied by means of
different devices: A resistor in parallel with a diode can be used to cause a voltage drop
across this element which is different with respect to the positive and the negative half-wave
(Lee & Habetler, 2003b). Other proposals provide the controlled bias by a soft starter, giving
rise to a similar effect (Zhang et at., 2008), or use a zero sequence voltage to estimate the
stator resistance of the machine (Jacobina et al., 2000).
Since the estimation of resistances is required for the tuning the control of inverter drives,
such methods are often integrated into the digital signal processing (DSP) software of the
drive. For this class of applications rotor and stator resistances may be focus of interest. In
this context observers (Jeon et al., 2002), models (Gao et al., 2008a) and neural networks
(Karanayil et al., 2007) are applied. In variable speed drives also signal injection methods
can be used to determine the resistances since the inverter can be seen as active voltage
source (Wu & Gao, 2006).

3.4 Detection of Rotor Eccentricity

Rotor eccentricities are usually classified by the terms static and dynamic eccentricity. A
visualization of these classes is depicted in Fig. 3. For static eccentricity the axis of rotation
of the rotor is displaced from geometric center of the stator. This eccentricity mode can be

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 171

caused by a misalignment of the bearings, the end shields or stator ovality. Since the air gap
and the air gap field are non-uniform in this case a so called unbalanced magnetic pull arises.
This pull acts in the direction of the minimum air gap. In the case of dynamic eccentricity
the axis of rotation of rotor is aligned with center of the stator, but the axis of rotation is
displaced from the rotor center. Dynamic eccentricity may be caused by a bent shaft,
mechanical resonances, bearing wear and static eccentricity. An unbalanced magnetic pull is
also present in this case (Dorrell 1996). In practice always mixed modes of eccentricity occur
(Faiz & Ojaghi 2008).
Mixed static and dynamic eccentricity gives rise to harmonic components with frequencies

1 s
f i[em]  f e (1  m ), (17)
P

in the stator currents, where P is the number of pole pairs and superscript e indicates the
reference to eccentricity related components and m  [1,2,3,...] is the ordinal number. Due to
the interaction of current and flux, power and torque specific components arise at

1 s
f te[ m ]  f e m . (18)
P

The slotting of the rotor also gives rise to additional frequencies

 1 s 
f iRe [ m]  f e (mR  c)  o ,
 
(19)
P

in the current, where c  [0,1] indicates static (zero) or dynamic (one) eccentricity, and o is
an odd integer number, o  [1,3,5,] .
For stationary operated induction machines, supplied by the mains, the eccentricity related
harmonic components (17)) or (19)) can be determined by means of a conventional Fourier
analysis (Thomson & Fenger 2001). Oscillating load effects require different approaches
since a Fourier analysis is not applicable any more. An alternative method utilizes the
negative sequence fault components of the stator current space phasor which is independent
of the actual loading of the machine (Wu et al., 2007).
Another approach strictly avoids time domain frequency analysis and is based on the
evaluation of (18) applied to instantaneous power (Kral et al., 2004b). A scheme of this
method is depicted in Fig. 12. In order to evaluate a quantity independent of the actual
rating of the machine, instantaneous power is divided by a reference power, which is, e.g.,
equal to the nominal power of the machine. A band pass filter extracts the frequency
components (18) and a phase locked loop (PLL) is used to track the filtered signal. Since
power is a real quantity, a Hilbert transform is applied to generate a complex phasor. The
angle  of the quantity associated to the output of the Hilbert transform is then used to
employ a data clustering technique similar to that one used for the Vienna Monitoring
Method in subsection 3.2. The data clustering averages the output of the filter in a spatial
domain to eliminate any frequencies other than the fault frequency.

www.intechopen.com
172 Fault Detection

Fig. 12. Method for the eccentricity detection based on instantaneous power; F = band pass
filter, PLL = phase locked loop, H = Hilbert transform, C = clustering technique, DFT =
discrete Fourier analysis

Eccentricities have mechanical origins and therefore fault detection based on vibration
signals leads to more direct – without the interaction of the magnetic field and the reaction
on the stator currents – and thus more significant signatures. A fault signature according to
(17) and (18) may also be caused by a mechanical mass imbalance of the rotating parts. This
case was investigated in (Kral et al., 2004c) with a non-rigid mounting of the machine. It
turned out that it is difficult to detect a mechanical problem by means of current signatures
whereas vibrational signatures very clearly indicate the issue. The result of this investigation
thus shows that the detection of mechanical problems by means of electrical signals is
limited.

4. Conclusions
In this chapter state of the art fault detection methods for permanent magnet and induction
machines are presented. The discussed methods rely on the evaluation of measured
voltages, currents and speed, respectively. Each kind of fault gives rise to a specific fault
pattern in either of the measured quantities. Advanced signal processing techniques and
neural network methods are used to isolate and assess the fault severity accordingly.
The field of rotating electric machine monitoring and diagnostics has seen major
advancements in the two decades. As a result of this, simple motor protection devices will
soon be replaced with sophisticated monitoring devices which will provide early indication
of impeding faults. The technology described in this chapter can be used to greatly improve
factory and process reliability and availability by reducing unscheduled downtime. The
potential cost savings are enormous. All of this can be achieved at little or no cost since all
the technology here relies only on the use of data from sensors that are all ready installed.
Virtually everything described here can be implemented with only software changes to
existing microprocessor-based protection relays.

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 173

The future will be even more exciting as advances are made in sensorless monitoring of
mechanical systems driven by electrical machines, motor bearings, and a wide variety of
motor types in a myriad of applications.

5. References
Antonino-Daviu, J. A.; Riera-Guasp, M.; Folch, J. R. & Palomares, M. P. M. (2006). Validation of a
new method for the diagnosis of rotor bar failures via wavelet transform in industrial
induction machines, IEEE Transactions on Industry Applications, vol. 42, no. 4, pp. 990-
996.
Bellini, A.; Fillipetti, F.; Franceschini, G.; Tassoni, C.; Passaglia, R.; Saottini, M. & Tontini, G.
(2000). ENELs Experience with On-Line Diagnosis of Large Induction Motors Cage
Failures, Conference Record of the IEEE Industry Application Conference, IAS.
Benbouzid, M. E. H. & Kliman, G. B. (2003). What stator current processing-based technique to
use for induction motor rotor faults diagnosis?, IEEE Transactions on Energy Conversion,
vol. 18, no. 2, pp. 238-244.
Bianchi, N., S. Bolognani and M. Zigliotto, (1996). Analysis of PM Synchronous Motor Drive
Failures during Flux Weakening Operation, in Proceedings of the 27th Annual IEEE
Power Electronics Specialist Conference - PESC96, Baveno, Italy, 23-27, pp. 1542-1548.
J.M. Cameron, W.T. Thomson and A.B. Dow. (1986). Vibration and Current Monitoring for
Detecting Airgap Eccentricity in Large Induction Motors, IEE Proceedings, vol. 133, pt.
B, no.3, pp. 155-163.
H. I. Choi and W. J. Williams, (1989). Improved time-frequency representation of multi-
component signals using exponential kernels, IEEE Trans. on Acoustics, Speech, and Signal
Processing, vol. 37, no. 6, pp. 862-871.
Cohen, L. (1989). Time-frequency distributions – A review, Proceedings of the IEEE, vol. 77, no. 7, ,
pp. 941-981.
Culbert, I. M. & Rhodes, W. (2007). Using Current Signature Analysis Technology to Reliably
Detect Cage Winding Defects in Squirrel-Cage Induction Motors, IEEE Transactions on
Industry Applications vol. 43, no. 2, pp. 422-428.
Dorrell, D. G. (1996). Calculation of Unbalanced Magnetic Pull in Small Cage Induction Motors
with Scewed Rotors and Dynamic Rotor Eccentricity, IEEE Transactions on Energy
Conversion, pp. 483-488.
Dorrell, D.G., Thomson, W.T., and Roach, S. (1997). Analysis of Airgap Flux, Current, and
Vibration Signals as a Function of the Combination of Static and Dynamic Airgap
Eccentricity in 3-Phase Induction Motors, IEEE Transactions on Industry Applications, vol.
33, no. 1, (January/February) pp. 24-34.
Faiz, J. & Ojaghi, M. (2009). Different indexes for eccentricity faults diagnosis in three-phase
squirrel-cage induction motors: A review, Mechatronics, vol. 19, no. 1, pp. 2 - 13.
Farag, S. F.; Bartheld, R. G. & Habetler, T. (1996). An Integrated on-Line Motor Protection
System, IEEE Industry Applications Magazine, vol. 2, no. 2, pp. 21-26.
Gao, Z.; Habetler, T. G.; Harley, R. G. & Colby, R. S. (2008a). A Sensorless Rotor Temperature
Estimator for Induction Machines Based on a Current Harmonic Spectral Estimation
Scheme, IEEE Transactions on Industrial Electronics, vol. 55, no. 1, pp. 407-416.

www.intechopen.com
174 Fault Detection

Gao, Z.; Colby, R. S.; Habetler, T. G. & Harley, R. G. (2008b). A Model Reduction Perspective on
Thermal Models for Induction Machine Overload Relays, IEEE Transactions on Industrial
Electronics, vol. 55, no. 10, (October 2008) pp. 3525-3534.
Gaylard, A.; Meyer, A. & Landy, C. (1995). Acoustic evaluation of faults in electrical machines,
Seventh International Conference on Electrical Machines and Drives, IEE (Conf. Publ. No. 412),
pp. 147-150.
Henao, H.; Demian, C. & Capolino, G. (2003). A frequency-domain detection of stator winding
faults in induction machines using an external flux sensor, IEEE Transactions on Industry
Applications, vol. 39, pp. 1272-1279.
Hurst, K. D. & Habetler, T. G. (1996). Sensorless speed measurement using current harmonic
spectral estimation in induction machine drives, IEEE Transactions on Power Electronics,
vol. 11, no. 1, pp. 66-73.
Jacobina, C. B.; Filho, J. E. C. & Lima, A. M. (2000). On-line estimation of the stator resistance of
induction machines based on zero-sequence model, IEEE Transactions on Power
Electronics, vol. 15, no. 2, pp. 346-353.
Jeon, S. H.; Oh, K. K. & Choi, J. Y. (2002). Flux observer with online tuning of stator and rotor
resistances for induction motors, IEEE Transactions on Industrial Electronics, vol. 49, no. 3,
pp. 653-664.
Karanayil, B.; Rahman, M. F. & Grantham, C. (2007). Online Stator and Rotor Resistance
Estimation Scheme Using Artificial Neural Networks for Vector Controlled Speed
Sensorless Induction Motor Drive, IEEE Transactions on Industrial Electronics, vol. 54, no.
1, pp. 167-176.
Kim, K. and Parlos, A. G. (2002). Induction motor fault diagnosis based on neuropredictors and
wavelet signal processing, IEEE/ASME Trans. Mechatronics, vol. 7, no. 2, , pp. 201-219.
Kliman, G.B., W.J. Premerlani, R.A. Koegl, D. Hoeweler, A New Approach to On-Line Turn Fault
Detection in ac Motors, Conference Record of the IEEE-IAS Annual Meeting, 1996, pp, pp.
687-693.
Kral, C.; Habetler, T. G.; Harley, R. G.; Pirker, F.; Pascoli, G.; Oberguggenberger, H. & Fenz, C. J.
M. (2003). A Comparison of Rotor Fault Detection Techniques with Respect to the
Assessment of Fault Severity, Symposium on Diagnostics for Electric Machines, Power
Electronics and Drives, SDEMPED, pp. 265-270.
Kral, C.; Habetler, T. G.; Harley, R. G.; Pirker, F.; Pascoli, G.; Oberguggenberger, H. & Fenz, C.
(2004a). Rotor Temperature Estimation of Squirrel-Cage Induction Motors by Means of
a Combined Scheme of Parameter Estimation and a Thermal Equivalent Model, IEEE
Transactions on Industry Applications, vol. 40, no. 4, pp, pp. 1049-1057, ISSN 0093-9994
Kral, C.; Habetler, T. G. & Harley, R. G. (2004b). Detection of Mechanical Imbalances of Induction
Machines Without Spectral Analysis of Time-Domain Signals, IEEE Transactions on
Industry Applications, vol. 40, no. 4, pp. 1101-1105, ISSN 0093-9994
Kral, C.; Fenz, C.; Plainer, M.; Pirker, F. & Pascoli, G. (2004c). Mechanical Imbalances---Test Bed,
Measurement, Detection Technique, International Conference on Electric Machines, ICEM.
Kral, C.; Kapeller, H.; Pirker, F. & Pascoli, G. (2005). Discrimination of Rotor Faults and Low
Frequency Load Torque Modulations of Squirrel Cage Induction Machines by Means of
the Vienna Monitoring Method, IEEE 36th Annual Power Electronics Specialists Conference,
PESC, Recife, Brazil, pp. 2861-2866.

www.intechopen.com
Condition Monitoring and Fault Detection of Electric Drives 175

Kral, C.; Gragger, J.; Pirker, F. & Pascoli, G. (2006). Sensorless Rotor Position Estimation through
Current Signature Analysis Based on Rotor Saliency Harmonics, International
Symposium on Power Electronics, Electrical Drives, Automation and Motion, SPEEDAM
2006.
Kral, C.; Kapeller, H. & Pirker, F. (2007). A Stator and Rotor Fault Detection Technique for
Induction Machines in Traction Applications for Electric or Hybrid Electric Vehicles,
The World Electric Vehicle Association Journal, WEVA, vol. 1, pp. 184-189.
Kral, C.; Pirker, F. & Pascoli, G. (2008a). The Impact of Inertia on Rotor Fault Effects - Theoretical
Aspects of the Vienna Monitoring Method, IEEE Transactions on power electronics, vol. 23,
no. 4, pp. 2136-2142.
Kral, C.; Haumer, A. & Bäuml, T. (2008b). Thermal Model and Behavior of a Totally-Enclosed-
Water-Cooled Squirrel-Cage Induction Machine for Traction Applications, IEEE
Transactions on Industrial Electronics, Vol. 55, No. 10, October 2008 3555, vol. 55, no. 10, pp.
3555-3565.
Le Roux, W., R.G. Harley and T.G. Habetler, (2002). Rotor Fault Analysis of a Permanent Magnet
Synchronous Machine, Proceedings of the 15th International Conference on Electrical
Machine (ICEM02) on CD, Brugge, Belgium, pp. 25-28.
Le Roux, W., R.G. Harley and T.G. Habetler, (2003). Detecting Rotor Faults in Permanent Magnet
Synchronous Machines, Proceedings of the 4th International Symposium on Diagnostics for
Electric Machines, Power Electronics and Drives on CD, Atlanta, Georgia, pp. 24-26.
Lee, S.-B., Habetler, T.G.; Harley, R.G.; and Gritter, D.J. (2002). "An evaluation of model-based
stator resistance estimation for induction motor stator winding temperature," IEEE
Transactions on Energy Conversion, vol. 17, no. 1, pp. 7-15.
Lee, S. B.; Tallam, R. M. & Habetler, T. G. (2003a). A robust, on-line turn-fault detection technique
for induction machines based on monitoring the sequence component impedance
matrix, IEEE Transactions on Power Electronics, Vol. 18, No. 3, pp. 865-872.
Lee, S. & Habetler, T. G. (2003b). An online stator winding resistance estimation technique for
temperature monitoring of line-connected induction machines, IEEE Transactions on
Industry Applications, vol. 39, no. 3, pp. 685-694.
Lee, Y.-K.; Habetler, T.G. (2006). A Stator Turn Fault Tolerant Strategy for Induction Motor
Drives in Safety Critical Applications, IEEE Power Electronics Specialists Conference, pp.
1–7.
S. L. Marple Jr. (1998). Time-frequency signal analysis: issues and alternative methods, in
Proceedings of the IEEE-SP International Symposium on Time-Frequency and Time-Scale
Analysis, pp. 329-332.
B.C. Mecrow, A.G. Jack, J.A. Haylock and J. Coles, (1996). Fault-Tolerant Permanent Magnet
Machine Drives, IEE Proceedings-Electric Power Applications, Vol. 143, No. 6, pp. 437-442.
Pal, S.K. (1991). Design Criteria for Brushless DC Motors with Hollow Rotor of Samarium Cobalt
for Applications above 25000 RPM in Vacuum, Proceedings of the 5th International
Conference on Electrical Machines and Drives, London, UK, pp. 115-120.
Rajagopalan, S., Le Roux, W., Habetler, T.G. and Harley, R.G. (2004). Diagnosis of potential rotor
faults in brushless DC machines, Second International Conference on Power
Electronics, Machines and Drives (PEMD 2004). vol. 2, pp. 668 – 673.

www.intechopen.com
176 Fault Detection

Rajagopalan, S.; Restrepo, J.A.; Aller, J.M.; Habetler, T.G.; Harley, R.G. (2005); Selecting time-
frequency representations for detecting rotor faults in BLDC motors operating under
rapidly varying operating conditions, Annual Conference of the IEEE Industrial Electronics
Society, IECON, pp. 6-10, 2005, (Nov 2005) pp. 2579 – 2584.
Stack, J. R.; Habetler, T. G. & Harley, R. G. (2006), 'Fault-signature modeling and detection of
inner-race bearing faults', IEEE Transactions on Industry Applications 42(1), 61-68.
Tallam, R. M.; Habetler, T. G.; Harley, R. G.; Gritter, D. J. & Burton, B. H. (2000). Neural Network
Based On-Line Stator Winding Turn Fault Detection for Induction Motors, Conference
Record of the IEEE Industry Application Conference, IAS.
Tallam, R. M.; Habetler, T. G. & Harley, R. G. (2003). Stator winding turn-fault detection for
closed-loop induction motor drives, IEEE Transactions on Industry Applications, vol. 39,
no. 3, pp. 720-724.
Tallam, R. M.; Lee, S. B.; Stone, G. C.; Kliman, G. B.; Yoo, J.; Habetler, T. G. & Harley, R. G. (2007).
A Survey of Methods for Detection of Stator-Related Faults in Induction Machines,
IEEE Transactions on Industry Applications, vol. 43, no. 4, pp. 920-933.
Thomson, W. T. & Ranking, D. (1987). Case History of Rotor Winding Fault Diagnosis in
Induction Motors, Conference Proceedings of Condition Monitoring, University College
Swanse, 798-819.
Thomson, W. T. & Fenger, M. (2001). Current Signature Analysis to Detect Induction Motor
Faults, IEEE Industry Applications Magazine, vol. 7, no. 4, pp. 26-34.
Wieser, R.; Kral, C.; Pirker, F. & Schagginger, M. (1999). On-Line Rotor Cage Monitoring of
Inverter-Fed Induction Machines by Means of an Improved Method, IEEE Transactions
on Power Electronics, vol. 14, no. 5, pp. 858-865.
Wolbank, T. M.; Loparo, K. A. & Wohrnschimmel, R. (2003). Inverter statistics for online
detection of stator asymmetries in inverter-fed induction motors, IEEE Transactions on
Industry Applications, no. 39, vol. 4, pp. 1102-1108.
Wu, L.; Habetler, T. & Harley, R. (2005). Separating Load Torque Oscillation and Rotor Fault
Effects in Stator Current-Based Motor Condition Monitoring, IEEE, International Electric
Machines and Drives Conference, 1889-1894.
Wu, Y. & Gao, H. (2006). Induction-motor stator and rotor winding temperature estimation using
signal injection method, IEEE Transactions on Industry Applications, vol. 43, no. 4, pp.
1038-1044.
Wu, L.; Habetler, T. & Harley, R. (2007). A Review of Separating Mechanical Load Effects from
Rotor Faults Detection in Induction Motors, IEEE International Symposium on Diagnostics
for Electric Machines, Power Electronics and Drives, The 6th, SDEMPED 2007, Cracow,
Poland.
B. Yazici and G. B. Kliman, (1999). An adaptive statistical time-frequency method for detection of
broken bars and bearing faults in motors using stator current, IEEE Trans. Industry
Applications, vol. 35, no. 2, pp. 442-452.
Zhang, P.; Lu, B. & Habetler, T. G. (2008). A Remote and Sensorless Stator Winding Resistance
Estimation Method for Thermal Protection of Soft-Starter-Connected Induction
Machines, IEEE Transactions on Industrial Electronics, vol. 55, no. 10, pp. 3611-3618.
Zhao Y.; Atlas, L. E.; Marks II, R. J. (1990). The use of cone-shaped kernels for generalized time-
frequency representations of nonstationary signals, IEEE Trans. on Acoustics, Speech, and
Signal Processing, vol. 38, no. 7, pp. 1084-1091.

www.intechopen.com
 Fault Detection
Edited by Wei Zhang

ISBN 978-953-307-037-7
Hard cover, 504 pages
Publisher InTech
Published online 01, March, 2010
Published in print edition March, 2010

In this book, a number of innovative fault diagnosis algorithms in recently years are introduced. These
methods can detect failures of various types of system effectively, and with a relatively high significance.

How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:

Christian Kral and Thomas G. Habetler (2010). Condition Monitoring and Fault Detection of Electric Drives,
Fault Detection, Wei Zhang (Ed.), ISBN: 978-953-307-037-7, InTech, Available from:
http://www.intechopen.com/books/fault-detection/condition-monitoring-and-fault-detection-of-electric-drives

InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
www.intechopen.com