1990 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO.

5, MAY 2011

A Simple Real-Time Fault Signature Monitoring Tool

for Motor-Drive-Embedded Fault Diagnosis Systems
Bilal Akin, Member, IEEE, Seungdeog Choi, Student Member, IEEE,
Umut Orguner, Member, IEEE, and Hamid A. Toliyat, Fellow, IEEE

Abstract—The reference frame theory constitutes an essential to motor failures. Therefore, fault diagnosis and condition
aspect of electric machine analysis and control. In this study, apart monitoring have been studied in the recent decade to pre-
from the conventional applications, it is reported that the reference vent costly interruptions due to motor faults [1]–[7]. As a
frame theory approach can successfully be applied to real-time
fault diagnosis of electric machinery systems as a powerful toolbox widely applied method, phase current analysis has received
to find the magnitude and phase quantities of fault signatures with much attention in search of providing a practical solution to
good precision as well. The basic idea is to convert the associated continuous monitoring and incipient fault detection [1]–[4].
fault signature to a dc quantity, followed by the computation of Although alternative methods such as vibration analysis with
the signal’s average in the fault reference frame to filter out the accelerometers and thermal analysis provide satisfactory results
rest of the signal harmonics, i.e., its ac components. As a natural
consequence of this, neither a notch filter nor a low-pass filter is [5]–[7], continual low-cost protection without the use of extra
required to eliminate fundamental component or noise content. sensors and hardware is always the most attractive method for
Since the incipient fault mechanisms have been studied for a long a greater market share.
time, the motor fault signature frequencies and fault models are Although the practical problems of utility-driven motor fault
very well-known. Therefore, ignoring all other components, the diagnosis are well known [3]–[6], the inverter-driven systems
proposed method focuses only on certain fault signatures in the
current spectrum depending on the examined motor fault. Broken need further analysis to overcome problems such as high
rotor bar and eccentricity faults are experimentally tested online noise floor of current spectrum, operating-point-dependent fault
using a TMS320F2812 digital signal processor (DSP) to prove the signature frequencies, transient-state dynamics of fault signa-
effectiveness of the proposed method. In this application, only tures, closed-loop and current regulator effects, etc. [8]–[15].
the readily available drive hardware is used without employing In the case of closed-loop control, the feedback and the current
additional components such as analog filters, signal conditioning
board, external sensors, etc. As the motor drive processing unit, regulator bandwidths can potentially yield erroneous results
the DSP is utilized both for motor control and fault detection unless their impacts are particularly taken into consideration
purposes, providing instantaneous fault information. The pro- during the analyses [8]–[10]. Furthermore, the impacts of the
posed algorithm processes the measured data in real time to avoid control parameters and physical modifications on decoupled
buffering and large-size memory needed in order to enhance the flux and torque components of stator current are different from
practicability of this method. Due to the short-time convergence
capability of the algorithm, the fault status is updated in each each other [10]. When the motor is run by an adjustable-speed
second. The immunity of the algorithm against non-ideal cases drive, the stator is excited at various frequencies depending on
such as measurement offset errors and phase unbalance is theo- the speed command and the applied load. This situation makes
retically and experimentally verified. Being a model-independent the fault signatures shift in the current spectrum and creates
fault analyzer, this method can be applied to all multiphase and difficulty in accurate fault signature tracking. Even though the
single-phase motors.
noise and various harmonic contents in the current spectrum
Index Terms—AC motor drive, broken rotor bar, condition of the inverter-fed motor degrade the accuracy of most analy-
monitoring, eccentricity, fault diagnosis, induction motor, refer- ses, among these, the low-order harmonics excite additional
ence frame theory, signal processing.
fault signatures which can yield more reliable fault decision
[11]. Various techniques are presented and applied successfully
I. I NTRODUCTION
to particular rotor fault diagnosis problems of field-oriented-

T HE indusrtry’s dependence on ac machines in critical

applications often results in very costly shutdowns due
controlled induction machine [12]–[15].
The line current information of inverter fed motors is readily
available for control and protection purposes. Thus, the new
Manuscript received June 2, 2009; revised September 29, 2009 and trend for low-cost protection is the motor-drive-integrated fault
March 16, 2010; accepted May 2, 2010. Date of publication June 10, 2010; diagnosis systems without using any external hardware. Even
date of current version April 13, 2011.
B. Akin is with Texas Instruments Incorporated, Dallas, TX 75243 USA
though numerous offline fault detection methods are reported
(e-mail: [email protected]). using costly hardware and high-level software, the implemen-
S. Choi and H. A. Toliyat are with the Department of Electrical and Com- tation of a low-cost real-time monitoring has still been a chal-
puter Engineering, Texas A&M University, College Station, TX 77840 USA
(e-mail: [email protected]). lenge due to the involved computational complexity. In order to
U. Orguner is with the Department of Electrical Engineering, Linköping achieve low-cost online monitoring, the diagnosis algorithms
University, 581 83 Linköping, Sweden. should be simple enough to be executed using industrial micro-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. processors in real time. Since thousands of data are processed,
Digital Object Identifier 10.1109/TIE.2010.2051936 and the number of data is a critical parameter for precision, the

0278-0046/$26.00 © 2010 IEEE

AKIN et al.: FAULT SIGNATURE MONITORING TOOL FOR MOTOR-DRIVE-EMBEDDED FAULT DIAGNOSIS SYSTEMS 1991

proposed solutions should avoid storing or buffering data. In In this paper, a simple motor-drive-integrated fault di-
order to examine the practicability, various techniques are im- agnosis method is proposed and implemented using the
plemented using industry-purposed microprocessors [15]–[21], core microprocessor of the inverter (Texas Instruments DSP
which is the pioneering step of low-cost drive-integrated fault TMS320F2812). The reference frame theory approach (RFTA)
diagnosis systems. is employed to measure the normalized amplitude of fault-
A model-based fault monitoring method avoiding frequency related harmonics. Instead of checking the whole current spec-
analysis is presented as a possible scheme that can be tailored trum or evaluating the current waveform roughly, each time,
to variable-speed drives in [15]. It is claimed in [16] that only a few fault-related harmonics are checked and measured to
the time–frequency distribution-based solutions are practical speed up the process and to lower the computational burden. As
enough for commercial implementations. In [16], after elimi- a basic tool, various reference-frame-theory-based applications
nating the fundamental component and high-order frequencies are reported in the literature, such as learning Park’s vector
using a switch capacitor filter, the stator current is processed pattern based on artificial neural network to discern healthy and
in real time by a digital signal processor (DSP) that runs a fault faulty patterns [22], finding deviation in an actual Concordia
diagnosis algorithm developed for nonstationary signals based pattern that is used to determine the types and magnitude
on Hilbert transformation and Zhao–Atlas–Marks distribution. of faults in drive systems and stator, respectively [23], [24],
In [17] and [18], a DSP is utilized both for motor control and obtaining negative-sequence stator-fault-related indices from
motor fault diagnosis based on lock-in detection and multiple the line current [25], and detecting negative-frequency rotor
reference frames, respectively. Another DSP implementation is asymmetry signatures at standstill based on complex fault
presented in [19] to detect mechanical load faults in induction signature vectors [26]. The scheme proposed in this paper is
motors during speed transients based on Wigner distribution. applicable to all kinds of motor faults at all operating points
Previously applied to sensorless motor control, a high-frequency even if only single-phase measurement is provided. Therefore,
signal injection technique is discussed in [20] as a possible it performs as a generic fault harmonic analyzer at user-defined
solution that can be integrated to fault diagnostic schemes. fault frequencies. Its performance does not depend on motor
The reference frame transformation-based algorithm adopted control strategy or machine parameters such as stator/rotor
in this paper is quite simple and efficient; therefore, it is directly resistances and machine inductances. As long as the motor
added into the main motor control subroutine (fast loop) in generates a periodic fault signature and the frequency of the
DSP program. Typically, four different fault signatures can be fault signature can be calculated or known, the RFTA can
tracked simultaneously in parallel without violating the CPU capture it both at the steady state and transients as it is motor
utilization limits and degrading the control performance. Be- model independent. The saturation and high temperature are
cause the stator frequency and shaft speed are mostly available physical facts basically affecting the parameters of the motor
as drive control parameters, the frequency of each fault signa- model. Since the proposed method does not depend on the
ture harmonic is updated dynamically using these parameters. motor parameters, it is not affected from saturation and high
These updated frequencies are utilized to synchronize the fault temperature. Neither analog nor digital filter is needed to elim-
reference frame and the fault vectors to monitor the motor inate high-energy content or various unpredictable operating-
condition not only at steady state but also during transients. point-dependent noise harmonics.
Furthermore, since each measured piece of data is processed in
real time, instead of buffering, the method does not suffer from II. A DAPTATION OF R EFERENCE F RAME T HEORY
memory shortage. The tests under nonideal conditions such as
offset, unbalance, etc., proved that they do not affect the results. A. Reference Frame Theory
The average of transformed signal in the fault reference frame The reference frame theory has successfully been applied
converges and settles down to a dc level quite fast; therefore, to various motor control schemes for more than a couple of
the diagnosis process can be finalized typically in 1 s. decades. Since its basics are well known, only a generalized
On the other hand, inverter-driven motor phase currents are version of it is given here briefly in (1) and (2). The commonly
influenced by many factors such as static and dynamic loading, used transformation is the poly-phase to orthogonal two-phase
motor speed transients, etc. Hence, the phase currents and mod- transformation. For the n-phase to two-phase case, it can be
ulated operating-point-dependent fault signatures are accepted expressed in the arbitrary reference frame as
as nonstationary signals [27]. However, it is well known that
Fourier transform techniques are not sufficient to represent non- [fxy ] = [T (θ)] · [f123,...,n ] (1)
stationary signals. When a nonstationary signal is transformed
where
into the frequency domain, most of the information about the
transient components of the signal is lost [29], [30], and hence, [T (θ)]
  
a hybrid method is proposed in [32]. Even the techniques such 2 cos(P θ/2) cos(P θ/2−ξ) · · · cos (P θ/2−(n−1)ξ)
as short-time Fourier transform, where a nonstationary signal is =
n sin(P θ/2) sin(P θ/2−ξ) · · · sin(P θ/2−(n−1)ξ)
divided into short pseudostationary segments, are not suitable (2)
for the analysis of signals with complex time–frequency charac-
teristics [27], [28]. During the experimental stage, it is observed with P being the number of poles, θ being the angle between the
that the partial use of these methods suffers from hardware stationary and rotating frames, and ξ being the electrical angle
limitations of low-cost real-time monitoring systems. between adjacent magnetic axes of the uniformly distributed
1992 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO. 5, MAY 2011

as fault reference frame. In other words, the fault reference

frame is a rotating reference frame which is synchronized to
the fault vector ifc in order to keep it stationary and convert the
vector and its orthogonal d−q projections to dc. The current
harmonic space vector in this fault reference frame is given by
îfc_dq = ifc_d + jifc_q
= îfc ej(θfc −ϕfc ) in the stationary frame
= îfc ej(ϕfc ) in the rotating (fault) frame (θ = θfc ) (4)
where θfc is the angle between the stationary reference frame
and the fault reference frame, ifc is the fault vector (vectorial
representation of the fault signature), and ϕfc is the phase angle
between the fault vector and fault reference frame. The phase-A
Fig. 1. Fault vector with other harmonic vectors in the stationary and rotating stator current can be modeled as
(fault) reference frames.
istator_A (t) = ifc cos(θfc −ϕfc )+inoise cos(θnoise −ϕnoise ) (5)
n-phase windings. One must note that the notations, indices,
and axes of the frames might change depending on how they where all the components in the stator current other than the
are defined by the user. In the literature, there are different fault signature are represented by the second term, including
representations of the reference frame theory, but the basics of fundamental component, all multiples of low-order harmonics,
all are the same. By judiciously choosing the reference frame, it dc offset, and the noise content. All we require from the noise
is possible to monitor motor faults, whose effects are reflected terms in (5) is that they do not have components at the fault
to the line current as detailed in the following section. Note that frequency. In practice, the phase current is composed of infinite
the fault reference frame modules in the software work sepa- harmonic vectors besides the fault components, where θfc =
rately and independently than the ones used for motor control. θnoise . If (5) is substituted into (1)–(4), all harmonics other than
the inspected one remain as ac in the fault reference frame, as
shown in the following equation:
B. Fault Component Analysis of Multiphase Systems
îstator_dq (t) = îfc e−jϕfc + înoise ej(θnoise −θfc −ϕnoise ) . (6)
The fault component space vector in the stationary reference      
stationary nonstationary
frame is defined as the complex quantity component components

îfc_αβ (t) = ifc_α (t) + jifc_β (t). (3) The average of these ac harmonics converges to zero and has
a negligible effect on the average after some sufficient time.
Complex current harmonic vectors describe a circular trajec- In other words, the reference frame synchronized to the fault
tory in the space vector plane as shown in Fig. 1. component shifts the frequency spectrum of the phase current
Therefore, a multiphase system in phase variables transforms by the frequency of fault component. This frame converts only
to a circular locus in the equivalent two-axis representation. the associated fault harmonic vector to a stationary vector at
In Fig. 1, the radius of the circle around the origin is the 0 Hz whose projection on orthogonal base vectors is dc, and
peak magnitude, and the vector rotation frequency is equal to the averages are nonzero in time. The resultant fault vector
the angular frequency of the inspected fault vector (vectorial modulation is normalized with respect to the fundamental
representation of the fault-related signal in the stator current). vector which is computed at the stator reference frame. The
Note that the drawings in Fig. 1 are exaggerated to explain ratio gives the relative magnitude of the fault harmonic as
explicitly the basics of the theory; indeed, the magnitude of (7), shown at the bottom of the page, where ifund_dk and
fundamental harmonic is several times higher than those of ifund_qk are the dq components of fundamental harmonic, θfund
all line and fault harmonics. If the rotating reference frame is is the angular position of the stator reference frame, and Ifund
defined where the axes are made to rotate at the same rate as and Ifc are the relative magnitudes of the fundamental and
the angular frequency of the inspected fault vector, a stationary fault harmonic vectors, respectively. In addition to the fault
current vector results, where its orthogonal components are dc harmonic magnitude calculation, the phase angle information
quantities. In this paper, the rotating reference frame is defined of associated harmonic vector can also be found using the

⎡  2  2  ⎤1/2

N 
N

⎢ ifc_dk +
1 1
ifc_qk ⎥
Ifc ⎢ N N ⎥
⎢ k=1 k=1
 θ=θfc ⎥
Ifund = 20 log ⎢
⎢   2   2 ⎥
⎥
dB (7)
⎣
N N
⎦
1
ifund_dk + N1 ifund_qk
N
k=1 k=1
θ=θfund
AKIN et al.: FAULT SIGNATURE MONITORING TOOL FOR MOTOR-DRIVE-EMBEDDED FAULT DIAGNOSIS SYSTEMS 1993

dq components obtained by the proposed technique as given

in the following equation:
⎛ ⎞

N
1
ifc_qk
⎜N ⎟
⎜ k=1 ⎟
ϕfc = tan−1 ⎜ θ=θfc ⎟ . (8)
⎝ 1 
N ⎠

ifc_dk
N
k=1 θ=θfc

C. Fault Component Analysis of a Single Phase

For further simplification, this method is reduced to a single
phase by assuming harmonic vector as pure imaginary or real.
The single-phase analysis has lesser computational burden and
provides implementation opportunity even if only one sensor
reading is available. Reducing the number of hardware com-
ponents like current sensors and interface circuit for signal
conditioning naturally yields greater reliability. Since the char- Fig. 2. Single-phase fault vector transformation.
acteristic features of the fault signatures are expected to be
identical in all three phases, the results are the same as that representations of vectors and frames in (9) are shown in Fig. 2.
of the multiphase analysis case. Although the averages of fault The fault component is decoupled into two as designated in the
frame dq components are different from the ones obtained for first part of (9), one of which is rotating synchronously with
multiphase, the normalized ratios remain the same. During the fault frame and the other component is rotating in the opposite
single-phase analysis, one of α or β components is set to zero, direction where the fault vector is oscillating on the real
and the other one is set to one of the three-phase current. axis of the stationary frame. When these two components are
Alternatively, setting both of them to the same phase quantity superimposed in the fault reference frame as given in the second
yields the same result as well. In order to depict single-phase part of (9), it is clearly seen that the first component of the
analysis, stationary frame real component is set to iα (t) = fault harmonic becomes dc and survives while the second one
ifc cos(θfc − ϕfc ) and iβ (t) = 0 as in the first case mentioned remains ac and vanishes after averaging as other nonstationary
earlier. After the reference frame transformation, the fault vec- harmonic content in the rotating fault reference frame.
tor synchronizes to the fault reference frame where θ = θfc as Similar to (6), the stator phase current is represented in the
in Fig. 2. Here, in order to explain the transformation visually fault reference frame as
as in the multiphase case, the rotating components of the single-
phase fault vector decoupled using Euler’s formula given as îstator_dq (t) = ifc e−jϕfc + ifc e−j(2θfc −ϕfc ) + înoise e−jθnoise .
⎛ ⎞      
ifc1 ifc2 stationary nonstationary
        component components
⎜ i(θ −ϕ ) ⎟ −iθ (10)
îfc_dq = ifc ⎜
⎝e
fc fc
2 + e−i(θfc −ϕfc ) 2⎟
⎠e
fc

Therefore, the mean of the projection of the stator current on the

    d-axis of the fault reference frame, including all the harmonics,
= ifc e−iϕfc 2 + e−i(2θfc −ϕfc ) 2 (9) is equal to the mean of the stationary component of the fault
vector on the d-axis, as shown in (11) at the bottom of the
where e−iϕfc /2 is the dc component, e−i(2θfc −ϕfc ) /2 is the ac page, where N is the number of processed data. Likewise, the
component, and ϕfc is the phase angle between the rotating mean of the fault harmonic on the q-axis can be calculated as
fault reference frame and the fault harmonic vector. The shown in (12) at the bottom of the page. Finally, the relative

  −jϕfc   j(θnoise (k)−θfc (k)−ϕnoise ) 

1 
N
e e−j(2θfc (k)−ϕfc ) e ej(θnoise (k)+θfc (k)−ϕnoise )
īstator_d =  ifc + + inoise +
N 2 2 2 2
k
ifc
−−−−−→ īfc_d = cos(ϕfc ) (11)
N →∞ 2

  −jϕfc   j(θnoise (k)−θfc (k)−ϕnoise ) 

1 
N
e e−j(2θfc (k)−ϕfc ) e ej(θnoise (k)+θfc (k)−ϕnoise )
īstator_q =  ifc + + inoise +
N 2 2 2 2
k
i
−−−−−→ īfc_q = fc sin(−ϕfc ) (12)
N →∞ 2
1994 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO. 5, MAY 2011

section, the impacts of the most often encountered problems on

the proposed technique are examined.

A. Offset Errors
First, the line current measurement offset error is studied,
which is mostly caused by either signal conditioning boards or
current sensing system. In this case, the stator phase current is
added an offset term ς
istator_A (t) = ifc cos(θfc − ϕfc )

+ inoise cos(θnoise − ϕnoise ) + ς. (16)

In the rotating frame (θ = θfc ), substituting (16) into (1) and

(2) for three-phase system analysis, the stator current vector is
written as
îstator_dq (t) = îfc e−jϕfc + înoise ej(θnoise −θfc −ϕnoise )
     
stationary nonstationary
component components

+ ςe−θfc . (17)
Fig. 3. Simulation of harmonic vector tracks when the reference frame is   
rotating at θ = θfc . (a) Fault component at 33.33 Hz. (b) First harmonic at nonstationary
component
60 Hz. (c) Third harmonic at 180 Hz. (d) Fifth harmonic at 300 Hz.
As mathematically derived in (10)–(12), the impact of noise
magnitude of the fault component in the fault reference frame is
component, including all harmonics, other than the fault-related
found as
! one does not affect the result, provided that the algorithm is run
ifc for a long enough time. As shown in (17), the dc component
|Ifc | = ī2fc_d + ī2fc_q = . (13)
2 turns out to be a vector rotating in the fault reference frame in
Note that the same procedure is repeated for the fundamental the opposite direction. Similar to other harmonics, the mean of
component in order to normalize the result. Thus, at the end, the its projections on the rotating axes converges to zero. Although
term N and constant coefficients are simplified, and an exact (16) and (17) are derived for the three-phase system, if (16) is
value of the fault component with respect to the fundamental substituted into (11)–(13), the same result can be obtained for
component is found. When the αβ components are taken as the single-phase case as well.
iα (t) = ia (t) and iβ (t) = ia (t), then the dc components in the The derivations are simulated both for the three- and single-
fault reference frame will converge to phase cases, as shown in Fig. 4, under practical assumptions.
Both the three- and single-phase stator current models have
ifc
īstator_d = (cos(ϕfc ) + sin(ϕfc )) (14) −40-dB fault signature at 144 Hz, 10% THD, 40-dB SNR, and
2 0.3-p.u. dc offset, as shown in Fig. 4(a) and (c). In the three-
ifc
īstator_q = (cos(ϕfc ) − sin(ϕfc )) (15) phase simulation, only one of the line current is added a dc
2
offset where the result would be the same as in the case when
where the relative ratio between the fault component and the all the phases are added. In Fig. 4(c) and (d), the time averaging
fundamental component magnitudes remains the same as cal- results show that the dq components of both methods converge
culated in the previous case and multiphase case. to the final average value quite fast even under various distort-
Simulation of typical harmonic vector tracks after single- ing impacts and dc offset simply by eliminating them in the
phase transformation given through (11)–(13) is shown in rotating fault frame. After repeating the same procedure for the
Fig. 3, where θ = θfc and stator current is composed of 1.0-p.u. fundamental component to normalize the fault component as
first harmonic, a number of 0.01-p.u. (−40-dB) low-order odd given in (7), both of the normalized fault signature magnitudes
harmonics, and −80-dB fault signal at 33.33 Hz. It is clear that are found as −40.087 dB, with and without offset. Note that the
only fault component harmonic has dc offset and that the rest accuracy of the final average values is further enhanced using
are located symmetrically around the origin with average values simple windowing techniques in order to get rid of the negative
converging to zero. Therefore, after a certain number of cycles, impacts of high-energy components.
the average of the transformed signal magnitude converges to
the magnitude of stationary vector, and the impacts of other B. Unbalanced Phase Currents
nonstationary components converge to zero.
The variations of the signal amplitudes mostly depend on the
parameter unbalance of the phases or supply unbalance. The
III. I MMUNITY OF RFTA AGAINST N ONIDEALITIES
relative effect of the phase unbalance on the fundamental
In the harsh industrial plant environment, it is highly possible component and the fault signature are proportional to each
to observe cases where the motors run under nonideal con- other. Because the proposed solution detects the faults based
ditions or measurement devices function inaccurately. In this on the normalized values, but not on the absolute magnitudes,
AKIN et al.: FAULT SIGNATURE MONITORING TOOL FOR MOTOR-DRIVE-EMBEDDED FAULT DIAGNOSIS SYSTEMS 1995

Fig. 5. (a) Unbalanced three-phase stator currents (in per unit): Phase A
has 1.5 times higher amplitude. (b) Time average of the fault signature dq
components when the frame is rotating at 144 Hz. (c) Same dq components
Fig. 4. (a) Three-phase stator currents (in per unit): Phase-A has 0.3-p.u.
as (b) when all three phases are balanced.
offset. (b) Time average of the fault signature dq components when the frame is
rotating at 144 Hz using the three-phase method. (c) Single-phase stator current TABLE I
with 0.3-p.u. offset. (d) Time average of the fault signature dq components ANALYSIS OF PROXIMATE HARMONICS
when the frame is rotating at 144 Hz using the single-phase method.

the “ratio” between the fault signature and the fundamental

component remains the same. This is valid both for single-
and three-phase analyses. Therefore, the proposed method is
not affected when the phase amplitudes are not the same. The
most common unbalanced stator conditions are unbalanced
source voltages and unbalanced loads [31]. In these cases,
the source voltages might be different as ega = egb = egc and
ega = egb = egc , or the applied voltages might be different due
to a series additional load as
2
vas = ega − ias Zs
3
1
vbs = egb + ibs Zs C. Proximity of the Harmonics
3
1 In some cases like broken bar fault analysis, the fault signa-
vcs = egc + ics Zs (18)
3 tures are in the close vicinity of the fundamental harmonic or
where vas , vbs , and vcs are the applied phase voltages and other rotor asymmetry components. Therefore, the resolution
Zs is the additional phase load that causes unbalanced load of the fault diagnosis tool should be high enough to correctly
situation in a Y-connected three-wired system. In addition to distinguish each harmonic component. In order to test the
this, the supply can be balanced, but the impedances might be RFTA, a number of neighboring harmonics are injected to the
different from each other. When the saturation is ignored, it is fundamental harmonic as shown in Table I, i.e., 54, 56, and
clear that all of the harmonic content proportionally changes in 58 Hz. Then, the proposed method is run at each frequency
each phase under all unbalanced conditions. The three-phase starting from 54 to 61 Hz, increasing the frame frequency by
transformation can be decoupled as a balanced three-phase one at each time. The simulation results prove that the proposed
vector and an additional single-phase vector causing unbalance, method is highly promising in differentiating the neighboring
both of which are detailed in the previous section. The resultant components, as shown in Table I, provided that sufficient data
averaged magnitude of each harmonic will be different from are used. Note that the amounts of decoupled dq components
the balanced condition; however, the normalized amplitudes vary depending on the phase angle of the injected harmonics.
remain constant. In Fig. 5(a), the unbalanced three-phase stator
current model has −40-dB fault signature at 144 Hz, 10% THD, D. Phase Delay
and 40-dB SNR. Phase-A current amplitude is intentionally
kept 50% higher than the other phases to simulate the unbal- Phase delay between the rotating frame and the inspected
anced three-phase system. After applying the procedure given harmonic or other harmonic components does not degrade the
through (1)–(7), the normalized fault components are calculated results. In Table I, the line currents are composed of different
as −40.247 and −40.087 dB for the unbalanced and balanced harmonics at different frequencies and phase delays. The phase
conditions, respectively. The simulation results show that the differences of injected harmonics change the amplitudes of the
proposed method can be applied to unbalanced systems to find dq components, as given in (6) and Table I, but do not affect the
normalized motor fault signature magnitudes reliably. magnitude of the examined component.
1996 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO. 5, MAY 2011

Fig. 7. Experimental setup.

with equal amount of delays running in parallel. If the experi-

enced system transients are too fast, either of these methods is
necessary in order to obtain acceptable precision.

F. Unfiltered Fundamental and High-Energy Components

The experimental and simulation tests verify that the res-
Fig. 6. (a) Phase A of stator current in transient state. (b) dq components olution of the fault detection can further be improved and
of the fundamental component. (c) dq components of the fault component. corrected using quite simple windowing techniques. Remember
(d) Time average of the normalized fault component.
that this technique operates on a finite-length time record in an
E. Transient-State Analysis attempt to estimate periodic harmonic vector magnitude, which
averages over all periods. When the current signal includes low-
So far, it is assumed that the rotor speed, fundamental stator order harmonics and particularly fundamental component with
frequency, and the load level are all constant. In practice, the high-energy content, the fault analysis suffers from these domi-
steady-state assumption in fault detection analysis is not valid nating components and results in slightly oscillatory waveforms
for many applications where constant speed cannot always as shown in Fig. 3(b) and (d). At the end of averaging, if all
be guaranteed (e.g., in wind generation or hybrid electric ve- ac components perfectly end up with an integer number of
hicle (HEV) drive cycles). The classical applications of fast periods, the final value would be free of error; however, this
Fourier transform (FFT) show an averaged frequency distrib- is an unrealistic case. The results can be improved either by
ution, and there is no time representation. Fluctuations in the processing more data or window weighting to minimize this
instantaneous frequency and amplitude as a result of machine error and ensure accuracy. The results obtained in Table I and
operation or loading effects are averaged out. Therefore, an the normalized fault component shown in Fig. 5(d) are obtained
accurate detection under transient conditions cannot easily be by means of a weighted window. As shown in Fig. 5(d),
accomplished. The transient state degrades the performance of zeroing the oscillatory behavior at the end, the normalized
RFTA as well, however, due to its particular features; the results average converges to the actual value with very high precision.
are superior to similar applications. Ignoring the rest of the Exhibiting robust features against the many nonidealities, in-
spectrum other than the inspected vector, RFTA provides fast accurate speed information would degrade the performance of
update opportunity which is an inevitable criterion for transient- the proposed method. This is because the fault signal in the
state analysis. Second, due to the readily available drive control new reference frame will not be stationary and the averaging
parameters such as stator frequency and rotor speed, one can will yield erroneous results. This issue is valid for many au-
continually update the operating-point-dependent fault signa- tocorrelation and discrete-Fourier-transform-based solutions as
ture frequencies on the order of microseconds and keep the well. A frequency offset search algorithm can partially solve
track of normalized magnitude during transients. this issue as long as the fault generates symmetric sidebands or
In order to simulate the transient response of the proposed its multiple harmonics to double check and justify the search
method, a stator current is modeled with exponentially ris- result. Otherwise, the relevancy of the findings to the inspected
ing amplitude and linearly increasing frequency as shown in fault signature cannot be guaranteed.
Fig. 6(a). A 0.01-p.u. fault component is injected to the stator
current, the frequency of which is five times higher than the
IV. E XPERIMENTAL R ESULTS
fundamental. The proportional increase in both fundamental
and harmonic currents during transient is a consistent assump- A number of identical induction motors are artificially mod-
tion with the transient model given in [31]. The transient-state ified to test broken rotor bar and eccentricity faults. The bro-
dq components of the fundamental and the fault components ken bar rotor has 4/44 broken bars, and the eccentric motor
obtained by (1)–(6) are shown in Fig. 6(b) and (c). The time is mechanically modified to form a mixed eccentricity with
average of the normalized fault component is shown in Fig. 6(d) 20.69% dynamic eccentricity and 41.37% static eccentricity. A
which shows the instantaneous update of (7). As shown in 3-hp induction motor is loaded by a dc generator and driven
Fig. 6(d), the normalized fault magnitude converges to the final by an inverter as shown in Fig. 7(a) and (b), respectively. The
value pretty fast and is accurately under transient condition inverter control [field-oriented control (FOC) or V /f ] and on-
as well. The update duration can be shortened by selecting a line fault diagnosis are implemented simultaneously by a 32-b
narrower window or multidata windows, following each other fixed-point 150-MHz TMS320F2812 DSP which has a 12-b
AKIN et al.: FAULT SIGNATURE MONITORING TOOL FOR MOTOR-DRIVE-EMBEDDED FAULT DIAGNOSIS SYSTEMS 1997

Fig. 8. Block diagram of FOC-driven motor fault diagnosis implementation.

embedded analog-to-digital converter (ADC). A 16-b ADC at

a 256-kHz SR760 FFT spectrum analyzer is used to monitor
the real-time current and voltage spectra to verify the online ex-
perimental findings. Several experiments are performed under
various conditions such as different rotor speeds, slip values,
load conditions, switching frequencies, sampling frequency,
and the number of data processed to examine the reliability of
the proposed technique.
Since the results obtained by RFTA are the actual magni-
tudes, proposing a particular fault index scheme is not needed.
The results are simply compared with a predetermined thresh-
old to check the severity of the fault. The shortcomings of
Fig. 9. DSP implementation timing.
constant threshold lookup table can be compensated using an
approximated adaptive threshold which is a function of the op-
erating point, i.e., the torque and the speed. In [16], an adaptive During the RFTA implementation, the number of processed
threshold that varies as a fixed percentage of the fundamental data is chosen to be the same as the sampling frequency to
frequency component (in this case chosen as 2%) is proposed synchronize the fault subroutine with a motor control algorithm
for an automatic detection. In order to come up with a complete as shown in Fig. 9. The number of data can typically be adjusted
dynamic behavior analysis of fault signatures, all the operating between 4000 and 20 000 depending on the resolution require-
points should be taken into consideration, including low-speed ments. When the motor is run from mains to get rid of insulated-
region, saturation region, etc., which is considered as future gate bipolar transistor switching frequency limit, and the DSP
work in this study. Furthermore, the closed- and open-loop is merely dedicated to the fault analysis, it is observed that
adaptive threshold patterns must be determined individually. the resolution enhancement efforts beyond 20 000 data yield
The load-dependent linear magnitude change assumption in the negligible benefit for this application. A predetermined number
case of utility-driven and open-loop inverter control is not valid of data in each data window help the user to simplify averaging
for the closed-loop applications. by downscaling each measured current sample without the need
of buffering. When the switching frequency is set to 20 kHz, it
takes 1 s to process 20 000 data, which is long enough to handle
A. DSP Implementation of RFTA
RFTA convergence time even at low frequencies where the
The ultimate goal of this study was to come up with a smart convergence takes longer time due to less numbers of averaging
motor drive that can both control the motor and report the periods.
incipient faults simultaneously using the drive-embedded DSP Instead of synchronizing the fault algorithm to switching
as shown in Fig. 8. Therefore, simple but robust methods are frequency, one can set ADC to continuous conversion and run
focused on in order not to overload the DSP during real-time consecutive fault diagnosis loops in each switching period. This
operations. can be implemented during the residual time after the motor
1998 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO. 5, MAY 2011

Fig. 10. DSP-based experimental results of open-loop V /f -controlled

inverter-driven motor sidebands. (a) Eccentricity signatures detected by RFTA. Fig. 11. DSP-based experimental results of field-oriented controlled inverter-
(b) FFT spectrum analyzer output of eccentric motor line current. (c) Broken driven motor sidebands. (a) Eccentricity signatures detected by RFTA.
bar signatures detected by RFTA. (d) FFT spectrum analyzer output of broken (b) FFT spectrum analyzer output of eccentric motor line current. (c) Broken
bar motor line current. bar signatures detected by RFTA. (d) FFT spectrum analyzer output of broken
bar motor line current.

control module to shorten the convergence time. Moreover, the calculated by DSP are very close to the FFT spectrum analyzer
number of data can further be reduced to get faster response in outputs at (29.79 Hz, 89.11 Hz) and (53.47 Hz, 65.19 Hz),
order to track the fault signature updates during the transients respectively. The errors between the RFTA results and spectrum
at the expense of resolution degradation. analyzer outputs are mostly less than |1| dB. The time spent
Owing to the low complexity of the proposed algorithm, a to process 20 000 data and find the amplitudes of sidebands
number of fault signatures can be estimated simultaneously. is 1 s. Depending on the resolution requirements and the
These signatures can belong to either various types of faults system control parameters, execution time might be shortened
or the same type of fault sidebands around the fundamental and or extended.
low-order harmonics. In this study, four different fault signa-
tures at various frequencies are tracked simultaneously during
the residual time following the motor control algorithm in each C. DSP-Based Fault Analysis of FOC-Controlled Motor
main interrupt subroutine (ISR) period. Since the incipient fault In Fig. 11, the same experiments are repeated, running the
propagation time is very long when compared to the order of motor with a closed-loop field-oriented control algorithm at
main ISR execution time, updating different types of fault status various operating points as detailed in Section IV-B. The eccen-
in each second is quite reasonable. The stator frequency can tricity test is implemented at no load, and the reference speed
either be found by current control module or set to the reference is set to wref = 0.66 p.u. where the sideband frequencies are
value depending on the control type, and the rotor speed can 20.02 and 59.57 Hz. The broken bar motor is tested under full
either be measured using an encoder or estimated to update fault load and wref = 0.6 p.u. where the sidebands are at 33.45 and
models. A 2048-pulse/rev. optical encoder is integrated to the 39.31 Hz. Similar to the V /f tests, the sideband errors are at an
motor shaft as shown in Fig. 7(b). Obviously, inaccurate speed acceptable level. These online experimental results confirm that
feedback yields less precise results due to the frequency offset; the proposed method can be adapted to real-time applications
however, an analysis of this issue is kept out of this paper’s successfully. In this study, it is observed that the most chal-
scope. The ADCs embedded in TMS320F2812 have 12 b which lenging case is encountered when the motor is controlled by
limits the signal sensing to −65 dB due to dominating quanti- FOC. One of the significant issues is higher noise floor which
zation errors as shown in Appendix B. For further resolution to threatens the small fault signature detection, and the other one is
detect the incipient fault at very early stages or bearing faults bandwidth-dependent error suppression impacts of the closed-
which generate very small fault signatures (typically between loop regulators. As long as FOC does not completely suppress
−70 and −90 dB), the system should be supported by integrated the fault signature, RFTA can detect the fault. However, the user
or embedded higher resolution ADCs. needs to adjust the severity factor and threshold when the motor
is run using current regulators.
B. DSP-Based Fault Analysis of V /f -Controlled Motor
D. Continuous Fault Monitoring in Time–Frequency Domain
In this experiment, the reference speed is set to ωref =
and Transient Analysis
0.99 p.u., and the motor is run at no load and full load for
eccentricity and broken bar tests, respectively. As shown in In addition to the previous sideband snapshots of eccentric
Fig. 10, both the eccentricity and broken rotor bar sidebands and broken bar faults, it is possible to trace the sideband
AKIN et al.: FAULT SIGNATURE MONITORING TOOL FOR MOTOR-DRIVE-EMBEDDED FAULT DIAGNOSIS SYSTEMS 1999

Fig. 13. DSP algorithm frequency update illustration of left eccentricity

sideband during the transient state.

stationary signals. On the other hand, RFTA keeps its focus

on only one component, the frequency information of which
is updated continuously in each sampling time, i.e., 50 μs in
this test. The updated frequency value is utilized to synchronize
the reference frame and the associated fault vector. Therefore,
even the motor excitation frequency or rotor shaft speed change,
the shifting fault signature in the spectrum is continuously
kept track of with very high frequency resolution as shown
in Fig. 13. Thus, the RFTA average contains more accurate
information regarding fault vectors.
This unique feature provides real-time tracking of fault sig-
nature components in the time–frequency domain. One can
keep the number of data relatively low, like 4000 instead
of 20 000, and update the fault status in every 0.2 s. Below
4000 data, it is observed that the resolution of the RFTA results
decreases drastically. However, if the motor has very low inertia
and the transient durations are very short, then, obviously,
RFTA cannot capture the fault signature during the transients.
The dynamic characteristics of right eccentricity sideband un-
der varying rotor speeds and load are traced experimentally by
Fig. 12. DSP-based experimental results of V /f -controlled inverter-driven
the DSP in real time as shown in Fig. 12(a) and (c), respec-
motor. (a) Normalized eccentricity sideband variation in time obtained by tively. During transients, the details of dynamic behaviors, loop
DSP running RFTA (at no load). (b) Related motor speed profile in per unit. impacts, related theoretical derivations, dynamic thresholding,
(c) Normalized eccentricity sideband variation in time obtained by DSP running
RFTA (wref = 0.4 p.u.). (d) Related motor line current vector in per unit.
etc., are considered as future work and, hence, out of this
paper’s scope.

variations in real time by means of RFTA. As shown in

V. C ONCLUSION
Fig. 12, the right eccentricity sideband amplitude variation
during speed and load change is monitored simply by updating This paper has presented the analytical validation of the
the result at the end of each data window. Instead of long- reference frame theory application to electrical motor fault di-
term periodic status checks, monitoring the motor continuously agnosis as an accurate amplitude and phase measurement tool.
and scheduling the needed maintenance prevent catastrophic The experimental test results are compared with FFT spectrum
failures. It is obvious that fault analysis based on stationary analyzer results to confirm the accuracy of this method. Its low
current is not realistic for most of the applications which have computational burden and immunity to nonideal cases enhance
dynamic duty cycle profile constituting transients such as HEV, its practical value for real-time applications. Being a motor-
wind generator, etc. In these cases, the motor runs at various model-independent structure, this technique can be applied to
operating points at different load–speed combinations for an fault diagnosis of other single- and multiphase ac motors. It
unknown time period. is experimentally shown that this simple diagnosis algorithm
During the transient state, the information provided by the can be embedded in the main control subroutine and run by the
FFT integral corresponds to all time instances. It means that motor drive processor in real time without affecting the control
no matter where in time the frequency appears, it will affect performance of the inverter. Therefore, it can even be added
the result of the integration equally. This is why the traditional to the existing motor drives at almost no cost, which is highly
application of FFT is not a satisfactory alternative for non- promising for future fault diagnosis products.
2000 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 58, NO. 5, MAY 2011

A PPENDIX [11] B. Akin, U. Orguner, H. A. Toliyat, and M. Rayner, “Low order PWM
inverter harmonics contributions to the inverter-fed induction machine
A. Some Fault-Related Identities fault diagnosis,” IEEE Trans. Ind. Electron., vol. 55, no. 2, pp. 610–619,
Feb. 2008.
The frequency of the eccentricity fault sidebands is given as [12] E. Serna and J. M. Pacas, “Detection of rotor faults in field oriented
   controlled induction machines,” in Conf. Rec. IEEE IAS Annu. Meeting,
1−s Oct. 2006, vol. 5, pp. 2326–2332.
feccentricity = f1 1 ± m [13] A. Bellini, C. Concari, G. Franceschini, and C. Tassoni, “Different pro-
p/2
cedures for the diagnosis of rotor fault in closed loop induction motors
= f1 ± m.fr , m = 1, 2, 3, . . . . (A1) drives,” in Proc. IEMDC, May 2007, vol. 2, pp. 1427–1433.
[14] S. M. A. Cruz and A. J. M. Cardoso, “Fault indicators for the diagnosis of
rotor faults in FOC induction motor drives,” in Proc. IEMDC, May 2007,
The frequency of the broken rotor bar fault sidebands is vol. 2, pp. 1136–1141.
given as [15] R. Wieser, C. Kral, F. Pirker, and M. Schagginger, “On-line rotor cage
monitoring of inverter-fed induction machines by means of an improved
fb = (1 ± 2ks)f1 , k = 1, 2, 3, . . . (A2) method,” IEEE Trans. Power Electron., vol. 14, no. 5, pp. 858–865,
Sep. 1999.
[16] S. Rajagopalan, T. G. Habetler, R. G. Harley, J. A. Restrepo, and
where m and k are integers, s is the per-unit slip, f1 is the J. M. Alle, “Non-stationary motor fault detection using recent quadratic
fundamental excitation frequency, fr is the mechanical rotation time-frequency representations,” in Conf. Rec. IEEE IAS Annu. Meeting,
frequency, and p is the number of poles. Oct. 2006, vol. 5, pp. 2333–2339.
[17] B. Akin, U. Orguner, H. A. Toliyat, and M. Rayner, “Phase-sensitive
detection of motor fault signatures in the presence of noise,” IEEE Trans.
Ind. Electron., vol. 55, no. 6, pp. 2539–2550, Jun. 2008.
B. Minimum Sensitivity Limit [18] S. M. A. Cruz, H. A. Toliyat, and A. J. M. Cardoso, “DSP implementation
of the multiple reference frames theory for the diagnosis of stator faults
The ADC resolution is given by in a DTC induction motor drive,” IEEE Trans. Energy Convers., vol. 20,
VRefHi − VRefLow no. 2, pp. 329–335, Jun. 2005.
Q= [19] M. Blodt, D. Bonacci, J. Regnier, M. Chabert, and J. Faucher, “On-line
2M − 1 monitoring of mechanical faults in variable-speed induction motor drives
3.0 V − 0 V using the Wigner distribution,” IEEE Trans. Ind. Electron., vol. 55, no. 2,
= = 0.732 mV/step
4096 − 1 pp. 522–533, Feb. 2008.
[20] F. Briz, M. W. Degner, A. Zamarron, and J. M. Guerrero, “Online stator
where Q is resolution in volts per step, VRefHi − VRefLow is the winding fault diagnosis in inverter-fed AC machines using high-frequency
signal injection,” IEEE Trans. Ind. Appl., vol. 39, no. 4, pp. 1109–1117,
full-scale voltage range for ADC input, and M is the ADC’s Jul./Aug. 2003.
resolution in bits. Therefore, the system cannot sense further [21] A. Kadir, T. Alukaidey, O. Al-Ayasrah, and R. Salman, “Embedded con-
details of the signal, and the absolute measurable lowest limit trol with predictive diagnostics algorithm of the induction machine drive
system,” in Proc. ACC, Jun. 2006, pp. 1080–1085.
is −62.7 dB while the normalized lowest limit is −65 dB. [22] H. Nejjari and M. E. H. Benbouzid, “Monitoring and diagnosis of in-
duction motors electrical faults using a current Park’s vector pattern
learning approach,” IEEE Trans. Ind. Appl., vol. 36, no. 3, pp. 730–735,
R EFERENCES May/Jun. 2000.
[1] G. B. Kliman, R. A. Koegl, J. Stein, R. D. Endicott, and M. W. [23] P. Gilreath and B. N. Singh, “A new centroid based fault detection method
Madden, “Noninvasive detection of broken rotor bars in operating induc- for 3-phase inverter-fed induction motors,” in Proc. IEEE Power Electron.
tion motors,” IEEE Trans. Energy Convers., vol. 3, no. 4, pp. 873–879, Spec. Conf., 2005, pp. 2664–2669.
Dec. 1988. [24] X. Tu, L. A. Dessaint, N. Fallati, and B. De Kelper, “Modeling and
[2] R. Schoen, T. Habetler, F. Kamran, and R. Bartfield, “Motor bearing dam- real-time simulation of internal faults in synchronous generators with
age detection using stator current monitoring,” IEEE Trans. Ind. Appl., parallel-connected windings,” IEEE Trans. Ind. Electron., vol. 54, no. 3,
vol. 31, no. 6, pp. 1274–1279, Nov./Dec. 1995. pp. 1400–1409, Jun. 2007.
[3] M. E. H. Benbouzid, “A review of induction motors signature analysis as [25] S. M. A. Cruz and A. J. M. Cardoso, “Multiple reference frames theory:
a medium for faults detection,” IEEE Trans. Ind. Electron., vol. 47, no. 5, A new method for the diagnosis of stator faults in three-phase induction
pp. 984–993, Oct. 2000. motors,” IEEE Trans. Energy Convers., vol. 20, no. 3, pp. 611–619,
[4] S. Nandi, M. Bharadwaj, and H. A. Toliyat, “Performance analysis of a Sep. 2005.
three-phase induction motor under mixed eccentricity condition,” IEEE [26] B. Akin, B. Ozturk, H. Toliyat, and M. Rayner, “DSP-based sensorless
Trans. Energy Convers., vol. 17, no. 3, pp. 392–399, Sep. 2002. electric motor fault diagnosis tools for electric and hybrid electric vehi-
[5] D. G. Dorrell, W. T. Thomson, and S. Roach, “Analysis of airgap flux, cle powertrain applications,” IEEE Trans. Veh. Technol., vol. 58, no. 5,
current, vibration signals as a function of the combination of static and pp. 2150–2159, Jun. 2009.
dynamic air gap eccentricity in 3-phase induction motors,” IEEE Trans. [27] B. Yazici and G. B. Kliman, “An adaptive statistical time-frequency
Ind. Appl., vol. 33, no. 1, pp. 24–34, Jan./Feb. 1997. method for detection of broken bars and bearing faults in motors using
[6] M. J. Devaney and L. Eren, “Detecting motor bearing faults,” IEEE stator current,” IEEE Trans. Ind. Appl., vol. 35, no. 2, pp. 442–452,
Instrum. Meas. Mag., vol. 7, no. 4, pp. 30–36, Dec. 2004. Mar./Apr. 1999.
[7] C. Riley, K. Lin, T. Habetler, and G. B. Kliman, “Stator current harmon- [28] H. Ocak, “Fault detection, diagnosis and prognosis of rolling element
ics and their causal vibrations: A preliminary investigation of sensorless bearings: Frequency domain methods and hidden Markov modelling,”
vibration monitoring applications,” IEEE Trans. Ind. Appl., vol. 35, no. 1, Ph.D. dissertation, School Elect. Eng., Case Western Reserve Univ.,
pp. 94–99, Jan./Feb. 1999. Cleveland, OH, 2004.
[8] R. Tallam, T. G. Habetler, and R. G. Harley, “Stator winding turn-fault [29] H. Douglas, P. Pillay, and A. K. Ziarani, “Broken rotor bar detection in
detection for closed-loop induction motor drives,” IEEE Trans. Ind. Appl., induction machines with transient operating speeds,” IEEE Trans. Energy
vol. 39, no. 3, pp. 720–724, May/Jun. 2003. Convers., vol. 20, no. 1, pp. 135–141, Mar. 2005.
[9] H. Xianghui, T. G. Habetler, and R. G. Harley, “Detection of rotor ec- [30] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of Electric Machinery.
centricity faults in a closed-loop drive-connected induction motor using Piscataway, NJ: IEEE Press, 1995.
an artificial neural network,” IEEE Trans. Power Electron., vol. 22, no. 4, [31] C.-M. Ong, Dynamic Simulation of Electric Machinery Using Matlab/
pp. 1552–1559, Jul. 2007. Simulink. Englewood Cliffs, NJ: Prentice-Hall, 1998.
[10] A. Bellini, F. Filippetti, G. Franceschini, and C. Tassoni, “Closed-loop [32] A. Lebaroud and G. Clerc, “Classification of induction machine faults
control impact on the diagnosis of induction motors faults,” IEEE Trans. by optimal time–frequency representations,” IEEE Trans. Ind. Electron.,
Ind. Appl., vol. 36, no. 5, pp. 1318–1329, Sep./Oct. 2000. vol. 55, no. 12, pp. 4290–4298, Dec. 2009.
AKIN et al.: FAULT SIGNATURE MONITORING TOOL FOR MOTOR-DRIVE-EMBEDDED FAULT DIAGNOSIS SYSTEMS 2001

Bilal Akin (S’03–M’08) received the B.S. and M.S. Hamid A. Toliyat (S’87–M’91–SM’96–F’08) re-
degrees in electrical engineering from Middle East ceived the B.S. degree in electrical engineering from
Technical University, Ankara, Turkey, in 2000 and Sharif University of Technology, Tehran, Iran, in
2003, respectively, and the Ph.D. degree in electrical 1982, the M.S. degree in electrical engineering from
engineering from Texas A&M University, College West Virginia University, Morgantown, in 1986, and
Station, in 2007. the Ph.D. degree in electrical engineering from the
From 2005 to 2007, he was an R&D Engineer University of Wisconsin, Madison, in 1991.
with Toshiba Industrial Division, Houston, TX. From In 1991, he was an Assistant Professor of electrical
2007 to 2008, he was a Postdoctoral Research As- engineering with the faculty of Ferdowsi Univer-
sociate with Texas A&M University. Since 2008, he sity of Mashhad, Mashhad, Iran. Since March 1994,
has been an Application Engineer with the C2000 he has been with the Department of Electrical and
Embedded Control Group, Texas Instruments Incorporated, Dallas. His re- Computer Engineering, Texas A&M University, College Station, where he
search interests are advanced control methods in motor drives, real-time fault is currently a Raytheon Endowed Professor. He has supervised more than
diagnosis of industrial systems, digital power management, and various DSP- 36 graduate students, published over 310 technical papers, where 92 papers are
based industrial applications. in IEEE T RANSACTIONS, and presented more than 50 invited lectures all over
the world. He is also an inventor and has ten issued and pending U.S. patents
in these fields. He is the author of DSP-Based Electromechanical Motion
Seungdeog Choi (S’08) received the B.S. degree in Control (CRC, 2003) and the coeditor of Handbook of Electric Motors—
electrical engineering from Chung-Ang University, 2nd Edition (Marcel Dekker, 2004). His main research interests and experience
Seoul, Korea, in 2004 and the M.S. degree in elec- include the analysis and design of electrical machines, variable-speed drives for
trical engineering from Seoul National University, traction and propulsion applications, fault diagnosis of electric machinery, and
Seoul, in 2006. He is currently working toward the sensorless variable-speed drives.
Ph.D. degree at Texas A&M University, College Dr. Toliyat is a member of Sigma Xi. He is an Editor of the IEEE
Station. T RANSACTIONS ON E NERGY C ONVERSION and was an Associate Editor
He was an R&D Engineer with LG Electronics, of the IEEE T RANSACTIONS ON P OWER E LECTRONICS. He is also the
Korea, from 2006 to 2007, and with Toshiba Inter- Chair of the Industrial Power Conversion Systems Department, IEEE-IAS.
national Cooperation, Houston, TX, from 2008 to He was the General Chair of the 2005 IEEE International Electric Machines
2010. His research interests include applied signal and Drives Conference in San Antonio, TX. He was the recipient of the 1996
processing, power electronics, and motor drive analysis with special emphasis and 2006 IEEE Power Engineering Society Prize Paper Awards and the 2006
on fault diagnosis. IEEE Industry Applications Society Transactions Third Prize Paper Award, the
prestigious Cyrill Veinott Award in Electromechanical Energy Conversion from
the IEEE Power Engineering Society in 2004, the TEES Fellow Award in 2004
and 2006, the Outstanding Professor Award in 2005 from Texas A&M, the
Umut Orguner (S’99–M’07) received the B.S.,
Distinguished Teaching Award in 2003, the E.D. Brockett Professorship Award
M.S., and Ph.D. degrees in electrical engineering
in 2002, the Eugene Webb Faculty Fellow Award in 2000, and the Texas A&M
from Middle East Technical University, Ankara,
Select Young Investigator Award in 1999 from Texas A&M University. He has
Turkey, in 1999, 2002, and 2006, respectively.
also received the Space Act Award from the National Aeronatical and Space
Between 1999 and 2007, he was a Teaching and
Administration in 1999 and the Schlumberger Foundation Technical Awards in
Research Assistant with the Department of Electrical
and Electronics Engineering, Middle East Techni- 2001 and 2000. He is a Professional Engineer in the State of Texas.
cal University. Since January 2007, he has been a
Postdoctoral Associate with the Division of Auto-
matic Control, Department of Electrical Engineer-
ing, Linköping University, Linköping, Sweden. His
research interests include estimation.