Anda di halaman 1dari 8

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 40, NO.

5, SEPTEMBER/OCTOBER 2004 1361

A New Algorithm for Transient Motor Current


Signature Analysis Using Wavelets
Hugh Douglas, Student Member, IEEE, Pragasen Pillay, Senior Member, IEEE, and Alireza K. Ziarani, Member, IEEE

AbstractA new algorithm is introduced for motor current


signature analysis of induction machines operating during tran-
sients. The algorithm is able to extract the amplitude, phase and
frequency of a single sinusoid embedded in a nonstationary wave-
form. The algorithm is applied to the detection of broken rotor
bars in induction machines during startup transients. The funda-
mental component of current, which varies in amplitude, phase,
and frequency, is extracted using the algorithm. The residual
current is then analyzed using wavelets for the detection of broken
rotor bars. This method of condition monitoring does not require
parameters such as speed or number of rotor bars, is not load
dependent and can be applied to motors that operate continuously
in the transient mode, e.g., wind generators or motor-operated
valves.
Index TermsAdaptive notch filter, broken rotor bars, condi-
tion monitoring, induction motors, starting transients, wavelets.

I. INTRODUCTION

T HE induction machine is essential in many industrial ap-


plications. It is therefore desirable to reduce downtime by
employing methods of machine condition monitoring. A widely
Fig. 1. Typical current spectrum of a fully loaded induction motor with broken
rotor bars.

used method of induction machine condition monitoring utilizes


the steady-state spectral components of stator quantities. These where
spectral components can include voltage, current and power and ;
can be used to detect broken rotor bars, bearing failures, air-gap supply frequency;
eccentricity, etc. Traditionally, these techniques have focused on machine slip;
the detection of faults during steady-state machine operation [1], number of poles.
[2]. The accuracy of these techniques depend on the loading of These frequencies form the stator current spectrum shown in
the machine, the assumption that the machine speed is constant, Fig. 1 and are present irrespective of the machines condition.
as well as the signal to noise ratio of the spectral components The presence of broken rotor bars is indicated by the difference
being examined. in amplitude between the fundamental and the left sideband. A
The rotor bar frequencies are determined by difference less than 50 dB is an indication of broken rotor bars
[3][7]. The amplitude of the left sideband frequency compo-
(1) nent of the fundamental frequency is proportional to the number
of broken rotor bars present [8]. The right sideband component,
, could also be used in monitoring fault severity. The
Paper IPCSD-04-037, presented at the 2003 Industry Applications Society importance of the stator current frequency components is well
Annual Meeting, Salt Lake City, UT, October 1216, and approved for publica- documented in [9][14].
tion in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial
Drives Committee of the IEEE Industry Applications Society. Manuscript sub- From (1) it is evident that the rotor bar frequencies are a func-
mitted for review June 1, 2003 and released for publication May 11, 2004. This tion of the machine slip. If the machine is unloaded, the slip
work was supported by the U.S. Navy Office of Naval Research. will be almost zero. The rotor bar frequencies will be masked
H. Douglas is with the Department of Electrical and Computer Engineering,
Clarkson University, Potsdam, NY 13699-5720 USA, on leave from the Depart- by the fundamental frequency and thus make detection difficult.
ment of Electrical Engineering, University of Cape Town, Rondebosch 7701, The only solution is, therefore, to heavily load the machine in
South Africa (e-mail: hdougla@eng.uct.ac.za). order to separate the frequencies. Overloading a machine is un-
P. Pillay is with the Department of Electrical and Computer Engineering,
Clarkson University, Potsdam, NY 13699 USA, and also with the University desirable since it reduces the machines operating lifetime and
of Cape Town, Cape Town, South Africa (e-mail: pillayp@clarkson.edu). is not generally under control of the operator. Accurate detec-
A. K. Ziarani is with the Department of Electrical and Computer En- tion therefore is difficult at light loading conditions. In addi-
gineering, Clarkson University, Potsdam, NY, 13699-5720 USA (e-mail:
pillayp@clarkson.edu; aziarani@clarkson.edu). tion many machines spend substantial periods at light loading
Digital Object Identifier 10.1109/TIA.2004.834130 conditions.
0093-9994/04$20.00 2004 IEEE
1362 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 40, NO. 5, SEPTEMBER/OCTOBER 2004

A fundamental disadvantage of the assumption of steady-


state speed in condition monitoring is that there are many ap-
plications where constant speed operation is not achieved, for
example, in wind generation or motor operated valves. In addi-
tion, the steady-state algorithms focus only on low slips, while
improved detection can be accomplished at high slips.
An alternate approach to the detection of broken rotor bars
would be to examine the starting transient of an induction ma-
Fig. 2. Illustration of the convergence mechanism of the gradient-descent
chine. The advantages would be that the transient has a high slip method employed to minimize a quadratic cost function.
and high signal-to-noise ratio, which implies that the spectral
components can be more easily separated. Loading does not af-
fect the amplitude of the transient during startup. The load only is the sinusoidal component of interest and is the totality
affects the duration time of the startup transient. This implies of the undesired components. The latter may in general be com-
that the detection can be done at low loading conditions unlike prised of noise, transient disturbances, and sinusoidal compo-
steady-state techniques. nents at frequencies other than . It is desired to obtain an es-
In [15] a method for the detection of broken rotor bars timate of , denoted by , out of the input signal .
and end ring faults, by examining the starting transient, is Least squares error between the input signal and the es-
proposed. The detection algorithm is based on steady-state timated sinusoidal signal embedded in , i.e., , is mini-
detection techniques by monitoring the sidebands of the current mized by the method of gradient descent. The cost function is
fundamental. Instead of using the Fourier transform to monitor defined as
the sideband frequencies, a Phase Vocoder [16] based on the
Short-Time Fourier Transform is used to produce a time-fre-
(4)
quency representation of the starting current. The shortcoming
of the Short-Time Fourier Transform is the fact that it analyzes
all the frequencies in a signal with the same window and causes where is the vector of parameters, e.g., the ampli-
resolution problems. The wavelet transform was introduced to tude, phase angle, and frequency, and is used to define the output
overcome the resolution problems encountered when analyzing (sinusoidal) signal. The gradient-descent algorithm provides a
transient signals using Fourier analysis. A detailed explanation method of adjusting the unknown parameter so that the cost
of wavelets applied to power engineering applications can be function tends to its minimum point. This method is based
found in [17][26]. on the idea of starting from an initial point and moving any un-
A challenge of transient analysis is the difficulty in trying to known parameter to the opposite direction of the variations of
analyze the complex transient startup current signal. This com- the cost function with respect to that parameter. If matrix
prises a nonstationary fundamental frequency as well as nonsta- is defined as in which , are
tionary frequencies associated with the rotor bars. The rotor bar real positive constants, then the gradient-descent method can be
frequencies are a function of the machine speed slip and change written as
as the machine runs up.
It is therefore desirable to be able to separate the fundamental
(5)
frequency from the rotor bar frequencies. Using a high-order
notch filter will not accomplish this goal because the funda-
mental frequency is not constant. A filter that actively tracks Fig. 2 illustrates the minimization process when achieved using
the changing amplitude, phase and frequency is needed to ex- the method of gradient descent in a simple case of one parameter
tract the fundamental from the transient. Once the fundamental and a quadratic cost function. This method is guaranteed to yield
frequency has been removed, the residual current can be exam- the desired solution if the cost function is globally quadratic in
ined using wavelets because the entire analysis is done in the parameters. Otherwise, i.e., if the form of the cost function is not
transient. quadratic, as is the case here, or not clearly described, a math-
ematical proof should be given to guarantee the convergence of
the solutions of the gradient-descent method (5) to the minimum
II. DESCRIPTION OF THE ALGORITHM
point of the cost function (4).
Let denote a signal comprising a sinusoidal component The output signal is defined as
and some undesired components

(2)

in which Formulating the algorithm accordingly using the parameter


vector , i.e., the amplitude, phase angle, and
(3) frequency of the desired component, results in the following
DOUGLAS et al.: NEW ALGORITHM FOR TRANSIENT MOTOR CURRENT SIGNATURE ANALYSIS USING WAVELETS 1363

where . First-order approximation for deriva-


tives is assumed in deriving these equations; in other words,
the time derivative of a generic quantity is approximated by
in discrete form. is the sampling time
and is the time index. It has been observed that using higher
order approximations does not present much improvement over
the first-order discretization presented here. This may be at-
tributed to the high robustness of the dynamics of the algorithm.
The state variables of the system of (6) can be envisaged in the
spherical coordinate system of . They can be rewritten
as

(7)

(8)
Fig. 3. Block diagram representation of the algorithm.
(9)

set of equations[27] which governs the dynamics of the Let where and
algorithm1: . The following theorem deals with the ex-
istence, uniqueness and stability of a periodic orbit for this dy-
namical system. The proofs are based on the Poincar theorem
and are presented in [28]. According to the Poincar theorem,
(6)
the behavior of the dynamical system near its periodic orbit can
be investigated using a discrete map. The fixed points of this
map correspond to the periodic orbits of the main dynamics and
The dot on top represents the differentiation with respect to their stability types are equivalent.
time and the error signal is . The state variables Theorem 1: Let where ,
, and directly provide instantaneous estimates of , and are real constants and is an arbitrary -pe-
the amplitude, phase angle and frequency of the extracted sinu- riodic bounded continuous function which has no frequency
soid, respectively. The totality of the undesired components and component at . With a proper choice of parameters
noise imposed on the sinusoidal component of interest is pro- , the dynamics described by (7)(9) have a unique peri-
vided by . The parameters , , and are positive num- odic orbit in space in a neighborhood of
bers which determine the behavior of the algorithm in terms of . This neighborhood is determined by the func-
convergence speed versus accuracy. This dynamics present an tion and the parameters to . Moreover, this periodic
algorithm which is capable of extracting a specified sinusoidal orbit is asymptotically stable. The periodic orbit coincides with
signal, estimating its amplitude, frequency, and phase, and ac- when is not present.
commodating variations in the amplitude, frequency and phase The theorem indicates that there is a unique periodic orbit to
of such a sinusoidal component. In practice, the frequency has which the system converges. This periodic orbit is located in a
a nominal value and varies around this nominal value. The neighborhood of the ideal desired component. The tighter this
second and third equations in (6) may be rearranged in terms of neighborhood is, the more accurately the desired component is
the deviation element . An implementation of this estimated. The extent of this neighborhood is determined by the
system is shown in Fig. 3 with three integrators for three state level of pollution and the step sizes , , and .
variables. The small values for step sizes and result in a refined
Fig. 3 shows implementation of the algorithm in the form of periodic orbit in a tight neighborhood. On the other hand, the step
the composition of simple blocks suitable for schematic software sizes determine the speed of the convergence to the solution of
development tools. Numerically, a possible way of writing the the differential equations. As well, if the parameters in the input
set of equations governing the present algorithm in discrete form, function (including amplitude and phase angle) vary with time,
which can be readily used in any programming language, is the desired solution will follow those variations provided that the
speed of the convergence to the solution, determined by the step
sizes, is sufficiently high. A tradeoff, therefore, exists between
the transient convergence speed and the steady state accuracy.
The dynamics of the algorithm presents a notch filter in the
sense that it extracts (i.e., lets pass through) one specific sinu-
1Strictly following the least squares error minimization using the method of soidal component and rejects all other components including
gradient descent results in a time-varying set of equations in which the time vari- noise. It is adaptive in the sense that the notch filter accommo-
able t is explicitly present in the equations. In the equations presented here, the dates variations of the characteristics of the desired output over
time variable t is replaced by a constant number. This replacement converts the time. The center frequency of such an adaptive notch filter is
time-varying system into a time-invariant system. The apparently arbitrary for-
mulation of the algorithm calls for mathematically rigorous justification which specified by the initial condition of frequency . In Fig. 3 this
is presented in [28]. initial value, , is explicitly shown for easy visualization.
1364 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 40, NO. 5, SEPTEMBER/OCTOBER 2004

Fig. 4. Illustration of the performance of the algorithm when both the


amplitude and frequency of the input signal undergo a jump of 100%.

In terms of the engineering performance of the system, this


indicates that the output of the system will approach a sinu-
soidal component of the input signal . Moreover, time vari-
ations of parameters in are tolerated by the system. Fig. 4
shows a snapshot of the performance of the algorithm when the
frequency and amplitude of the input signal jump from 50 to
100 Hz and 1 to 2, respectively. Initially, the periodic orbit is
a circle with unit radius which lies on the horizontal plane of
Hz, then it flows to another circle with radius 2 which
lies on the plane of Hz.
One of the issues that needs to be considered when using
the algorithm is setting of its parameters , , and . The
values of the parameters , , and determine the conver-
gence speed versus error compromise. Specifically, parameter
controls the speed of the transient response of the algorithm
with respect to variations in the amplitude of the extracted sinu-
soid. Speed is traded off by the steady-state error. As long as the
frequency of the input signal is close to its nominal value (e.g.,
60 Hz), this tradeoff does not introduce a significant constraint.
As the frequency of the input signal deviates from its nominal
value, the algorithm introduces more significant tradeoff be-
tween the speed and steady-state error; for example, within a
range of variations off nominal frequency, the algorithm
can be adjusted to catch up a 100% step change in the ampli-
tude of the desired sinusoid within 3 cycles with less than 2%
steady-state error. Parameters and mutually control the
speed of the transient response of the algorithm with respect to
variations in the frequency of the extracted sinusoid.
The larger the value of parameter is chosen, the faster the
convergence is achieved in tracking the phase (or frequency)
variations over time. The cost is higher steady state error. Ad-
justment of the third tuning parameter, , is interdependent on
the adjustment of . As a simple rule of thumb, one may pick
a value for parameter based on the potential frequency drift
in the input signal (hence, the desired frequency tracking speed) Fig. 5. Response of the algorithm to a step change in the frequency of the input
first, and choose the value of such that the product be- signal. The sinusoidal component and all its parameters are tracked adaptively
over time.
comes of the same order of for a balanced speed in terms of
tracking the amplitude and total phase (or frequency).
In the simulations presented in this section, Matlab Simulink Fig. 5 shows the performance of the algorithm for an example in
computational software is used as the main computational tool. which the frequency of the input signal undergoes a step change
DOUGLAS et al.: NEW ALGORITHM FOR TRANSIENT MOTOR CURRENT SIGNATURE ANALYSIS USING WAVELETS 1365

Fig. 7. Startup current transient.


Fig. 6. Illustration of the performance of the algorithm in the extraction of the
amplitude and phase of the fundamental component of a distorted sinusoid. Top:
input signal (solid), extracted fundamental (dashed), and its amplitude (dotted).
Bottom: actual (solid) and extracted (dashed) phase angles.

of 10%. It is observed that the variations are effectively tracked


with a transient lasting just a few cycles. Values of the param-
eters are chosen to be , , and
for this simulation.
The algorithm is particularly efficient for real-time appli-
cations. To illustrate the real-time nature of the algorithm,
an example having a distorted sinusoid as the input signal is
considered. Fig. 6 shows the performance of the algorithm in
the extraction of the amplitude and phase of the fundamental
component of this distorted sinusoid. The distortion in this case
is of a harmonic form with 30% of the third and 10% of the
tenth harmonic. It is observed that the amplitude and phase
of the fundamental component are estimated almost within Fig. 8. Plot of the current vector.
two cycles. Unlike the Fourier-based techniques in which only
estimates of the amplitude and constant phase are computed,
the fundamental component itself is instantly generated and is
available in real time. This is due to the fact that the algorithm
instantly generates the total phase rather than .

III. APPLICATION OF THE ALGORITHM


The measured startup current transient of an 11-kW induc-
tion motor is shown in Fig. 7. Before implementing the algo-
rithm, the individual measured line currents are transformed
into a single rotating current vector as shown in Fig. 8 [29],
[30]. The reason for combining the three line currents is that
no two starting transients are exactly the same. The individual
line currents will differ depending on instantaneous value of the
Fig. 9. Time-domain representation of the current vector.
line voltages at startup. When comparing the three line currents
under different machine health conditions, it will be easier to
compare a single rotating vector than three line currents that are the input. The resulting waveform shown in Fig. 12 has informa-
not similar. tion relating to the health of the machine including bad bearings,
The vector is then transformed into the time domain and used broken rotor bars, etc.
as an input to the extraction algorithm. The algorithm estimates The algorithm takes a few cycles to converge to the amplitude
the frequency, amplitude and phase of the nonstationary fun- and frequency of the fundamental. This is shown in Figs. 10 and
damental as shown in Figs. 911. The fundamental component 12. As a result when the estimated fundamental is subtracted
(which varies with magnitude, frequency and phase) can be ex- from the original waveform, the algorithms output between
tracted with this algorithm. This estimate is then subtracted from 00.4 s should be discarded to allow for convergence.
1366 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 40, NO. 5, SEPTEMBER/OCTOBER 2004

Fig. 10. Estimated fundamental startup current of the algorithm.

Fig. 11. Frequency of the algorithm.

Fig. 13. Wavelet decomposition levels D9 of a healthy machine loaded


30%100%.

Fig. 12. Startup current after extraction of the fundamental.

Fig. 12 shows the estimated frequency of the fundamental. An


accurate estimate of the frequency is only available after 0.4 s.

IV. DETECTION OF BROKEN ROTOR BARS


The algorithm was used to detect broken rotor bars in a 1/2-hp
induction motor. Two identical rotors were used in this experi-
ment except that one had a broken rotor bar. The same bearings
and stator was used in order to minimize their influences on the
startup transients. The machine was tested under loading con-
ditions varying from 30% to 100% to determine if this method
of detection could be successful and independent of the loading
conditions.
The current signals were captured using a PC-based data ac- Fig. 14. Wavelet decomposition levels D9 of a damaged machine loaded
quisition system. The nonstationary fundamental was removed 30%100%.
DOUGLAS et al.: NEW ALGORITHM FOR TRANSIENT MOTOR CURRENT SIGNATURE ANALYSIS USING WAVELETS 1367

using the system in Fig. 3 and implemented using Matlabs [9] M. E. H. Benbouzid, A review of induction motors signature analysis
Simulink. The residual currents were analyzed using Matlabs as a medium for faults detection, IEEE Trans. Ind. Electron., vol. 47,
pp. 984993, Oct. 2000.
Wavelet Toolbox. [10] W. Deleroi, Squirrel cage motor with broken bar in the rotor- physical
The discrete wavelet transform, using Daubechies 8 wavelet, phenomena and their experimental assessment, in Proc. Int. Conf. Elec-
trical Machines, Budapest, Hungary, 1982, pp. 767770.
was then applied to the residual current vector. Figs. 11 and 12 [11] H. A. Toliyat, Condition monitoring and fault diagnosis of electrical
show the ninth scale detail coefficients of both healthy and dam- machinesA review, in Proc. Int. Conf. Industry Applications, vol. 1,
aged machine under various loads. By inspection of Figs. 13 and Seattle, WA, 1999, pp. 197204.
[12] F. Filippetti, AI techniques in induction machines diagnosis including
14, two dominant features are present that characterize the condi- the speed ripple effect, IEEE Trans. Ind. Applicat., vol. 34, pp. 98108,
tion of the machine. The first feature is found between samples 8 Jan./Feb. 1998.
and 13 of all the loading conditions. This feature is present in both [13] M. Haji and H. A. Toliyat, Pattern recognitionA technique for induc-
tion machines rotor fault detection, in Proc. IEEE Electric Machines
the healthy and damaged machine. The second feature found be- and Drives Conf., Cambridge, MA, 2001, pp. 899904.
tween samples 45 and 53 is only present in the case of the dam- [14] P. Pillay and Z. Xu, Motor current signature analysis, in Conf. Rec.
aged machine. The second feature can be used to discriminate IEEE-IAS Annu. Meeting, vol. 1, San Diego, CA, 1996, pp. 587594.
[15] J. F. Watson and S. Elder, Fault detection in induction motors as a result
between a healthy and a damaged machine. An automated fault of transient analysis, in Proc. Fourth Int. Conf. Electrical Machines and
detection analyzer is envisioned based on this algorithm. Drives, 1989, pp. 182186.
[16] R. W. Schafer and L. R. Rabiner, Design and analysis of a speech anal-
ysis-synthesis system based on short time Fourier analysis, IEEE Trans.
V. CONCLUSION Audio Electroacoust., vol. 21, pp. 165174, June 1973.
[17] D. C. Robertson, O. I. Camps, and J. S. Mayer, Wavelets and power
A new algorithm for use in transient motor current signa- system transients: Feature detection and classification, in Proc. Int.
ture analysis has been introduced and applied to the detection Symp. Optical Engineering in Aerospace Sensing, vol. 2, 1994, pp.
474487.
of broken rotor bars in induction machines. The algorithm is [18] S. Santoso, E. J. Powers, and W. M. Grady, Power quality assessment
able to extract a single nonstationary sinusoid embedded within via wavelet transform analysis, IEEE Trans. Power Delivery, vol. 11,
a nonstationary waveform. pp. 924930, Apr. 1996.
[19] P. Pillay and A. Bhattacharjee, Application of wavelets to model short-
This is then applied to the transient inrush current of an induc- term power system disturbances, IEEE Trans. Power Syst., vol. 11, pp.
tion motor and a wavelet analysis is conducted on the balance 20312037, Oct. 1996.
of the current. [20] A. W. Galli, G. T. Heydt, and P. F. Riberio, Exploring the power of
wavelet analysis, IEEE Comput. Applicat. Power, vol. 9, pp. 3741,
Although results presented here are unique to the machine Oct. 1996.
being diagnosed, the methodology applied to extracting the tran- [21] P. Pillay, P. Ribeiro, and Q. Pan, Power quality modeling using
sient fundamental current and analysis of the residual currents wavelets, in Proc. Int. Conf. Harmonic and Quality of Power, Oct.
1996, pp. 625631.
is universal and can be applied to any machine. [22] A. Gaouda and M. Salama, Wavelet-based signal processing for distur-
The analysis clearly shows that the broken rotor bar can be bance classification and measurement, Proc. IEEGeneration, Trans.,
detected using measured transient inrush currents. This method Distrib., vol. 149, no. 3, pp. 310318, May 2002.
[23] P. S. Meliopoulos and C. H. Lee, Wavelet based transient analysis,
can be used for standard induction motors during startups as IEEE Transact. Power Delivery, vol. 15, pp. 114121, Jan. 2000.
well as machines that operate predominantly in the transient like [24] I. Daubechies, Orthonormal bases of compactly supported wavelets,
Commun. Pure Appl. Math., vol. 41, pp. 909996, 1988.
wind generators or motor operated valves. [25] P. F. Ribeiro, Wavelet transform: An advanced tool for analyzing non-
stationary harmonic distortions in power systems, in Proc. IEEE Int.
ACKNOWLEDGMENT Conf. Harmonics in Power Systems, Sept. 1994, pp. 141149.
[26] S. G. Mallat, A theory for multiresolution signal decomposition: The
The authors acknowledge the University of Cape Town Power wavelet representation, IEEE Trans. Pattern Anal. Machine Intell., vol.
7, pp. 674693, July 1989.
Engineering Group for assistance with this project. [27] A. K. Ziarani and A. Konrad, A nonlinear adaptive method of elimina-
tion of power line interference in ECG signals, IEEE Trans. Biomed.
Eng., vol. 49, pp. 540547, June 2002.
REFERENCES [28] M. Karimi-Ghartemani and A. K. Ziarani, Periodic orbit analysis of
[1] G. B. Kliman and J. Stein, Methods of motor current signature anal- two dynamical systems for electrical engineering applications, J. Eng.
ysis, Elect. Mach. Power Syst., vol. 20, no. 5, pp. 463474, Sept. 1992. Math., vol. 45, no. 2, pp. 135154, 2003.
[2] W. Deleroi, Broken bars in squirrel cage rotor of an induction motor [29] A. J. M. Cardoso, Computer-aided detection of airgap eccentricity in
Part 1: Description by superimposed fault currents, Arch. Elektrotech., operating three-phase induction motors by Parks vector approach,
vol. 67, pp. 9199, 1984. IEEE Trans. Ind. Applicat., vol. 29, pp. 897901, Sept./Oct. 1993.
[3] R. Hirvonen, On-line condition monitoring of defects in squirrel cage [30] M. E. H. Benbouzid, Monitoring and diagnosis of induction motors
motors, in Proc. Int. Conf. Electrical Machines, vol. 2, Paris, France, electrical faults using a current Parks vector pattern approach, IEEE
1994, pp. 301308. Trans. Ind. Applicat., vol. 36, pp. 730735, May/June 2000.
[4] M. E. H. Benbouzid and M. Vieira, Induction motor fault detection
and localization using stator current, advanced signal processing tech-
niques, IEEE Trans. Power Electron., vol. 14, pp. 1422, Jan. 1999.
[5] S. L. Ho, W. L. Chan, and H. W. Leung, Application of statistical signal
processing for condition monitoring of rotor faults in induction motors,
in Proc. Int. Conf. Electrical Machines, Oxford, U.K., 1993, pp. 97102. Hugh Douglas (S03) received the Bachelors and
[6] G. B. Kliman, What stator current processing-based technique to use Masters degrees from the University of Cape Town,
for induction motor rotor faults diagnosis?, IEEE Trans. Energy Con- Rondebosch, South Africa, in 1997 and 2000, respec-
version, vol. 18, pp. 238244, June 2003. tively. He is currently working toward the Ph.D. de-
[7] W. T. Thomson and I. D. Stewart, On-line current monitoring for fault gree in electrical engineering at Clarkson University,
diagnosis in inverter fed induction motors, in Proc. Third Int. Conf. Potsdam, NY.
Power Electronics and Variable Speed-Drives, London, U.K., 1994, pp. He is a faculty member at the University of Cape
6673. Town, where he specializes in the fields of power
[8] C. Hargis, The detection of rotor defects in induction motors, in Proc. electronics, machines, and drives. His research
IEE Int. Conf. Electrical Machines, Design and Application, London, interests include signal processing applied to power
U.K., 1982, pp. 216220. applications.
1368 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 40, NO. 5, SEPTEMBER/OCTOBER 2004

Pragasen Pillay (S84M87SM92) received Alireza K. Ziarani (S99M02) received the B.Sc.
the Bachelors degree from the University of degree in electrical and communication systems engi-
Durban-Westville, Durban, South Africa, in 1981, neering from Tehran Polytechnic University, Tehran,
the Masters degree from the University of Natal, Iran, in 1994, and the M.A.Sc. and Ph.D. degrees in
Durban, South Africa, in 1983, and the Ph.D. de- electrical engineering from the University of Toronto,
gree from Virginia Polytechnic Institute and State Toronto, ON, Canada. in 1999 and 2002, respectively.
University, Blacksburg, in 1987. In 2002, he joined the Department of Electrical
From January 1988 to August 1990, he was with and Computer Engineering, Clarkson University,
the University of Newcastle-upon-Tyne, U.K. From Potsdam, NY, where he is presently an Assistant
August 1990 to August 1995, he was with the Univer- Professor. He is also the co-founder of Signamatic
sity of New Orleans. Currently, he is with Clarkson Technologies, Toronto, ON, Canada. His research
University, Potsdam, NY, where he is a Professor in the Department of Elec- interests include nonlinear adaptive signal processing, biomedical engineering,
trical and Computer Engineering and holds the Jean Newell Distinguished Pro- embedded systems design, and the theory of differential equations.
fessorship in Engineering. He is also an Adjunct Professor at the University of
Cape Town, Cape Town, South Africa. His research and teaching interests are
in modeling, design, and control of electric motors and drives for industrial and
alternate energy applications.
Dr. Pillay is a member of the IEEE Power Engineering, IEEE Industry Appli-
cations (IAS), IEEE Industrial Electronics, and IEEE Power Electronics Soci-
eties. He is a member of the Electric Machines Committee and Past Chairman of
the Industrial Drives Committee of the IAS and Past Chairman of the Induction
Machinery Subcommittee of the IEEE Power Engineering Society. He currently
chairs the Awards Committee of the IAS Industrial Power Conversion Depart-
ment. He has organized and taught short courses in electric drives at IAS Annual
Meetings. He is a Fellow of the Institution of Electrical Engineers, U.K., and a
Chartered Electrical Engineer in the U.K. He is also a Member of the Academy
of Science of South Africa. He was also a recipient of a Fulbright Scholarship.

Anda mungkin juga menyukai