You are on page 1of 8

-

CONDITION MONITORINGAND FAULT


DIAGNOSIS
OF ELECTRICAL
MACHINESA

REVIEW
Subhasis Nandi

Hamid A. Toliyat

Student Member, IEEE

Senior Member, IEEE

Electric Machines & Power Electronics Laboratory


Department of Elecuical Engineering
Texas A&M University
College Station, TX 77843-3128
Abstract-Recently,research has picked up a fervent pace in the
area of fault diagnosis of electrical machines. Like adjustable
speed drives, fault prognosis has become almost indispensable.
The manufacturers of these drives are now keen to include
diagnostic features in the software to decrease machine down
time and improve salability. Prodigious improvement in signal
processing hardware and software has made this possible.
Primarily, these techniques depend upon locating specific
harmonic components in the line current, also known as motor
current signature analysis (MCSA). These harmonic
components are usually different for different types of faults.
However with multiple faults or different varieties of drive
schemes, MCSA can become an onerous task as different types
of faults and time harmonics may end np generating similar
signatures. Thus other signals such as speed, torque, noise,
vibration etc., are also explored for their frequency contents.
Sometimes, altogether different techniques such as thermal
measurements, chemical analysis, etc., are also employed to find
out the nature and the degree of the fault. Going by the present
trend, human involvement in the actual fault detection decision
making is slowly being replaced by automated tools such as
expert systems, neural networks, fuzzy logic based systems; to
name a few. It is indeed evident that this area is vast in scope.
Hence, keeping in mind the need for future research, a review
paper describing different types of faults and the signatures they
generate and their diagnostics schemes, will not be entirely out
of place. In particular, such a renew helps to avoid repetition of
past work and gives a birds eye-view to a new researcher in this
area.

1. INTRODUCTION

The history of fault diagnosis and protection is as archaic


as the machines themselves. The manufacturers and users of
electrical machines initially relied on simple protections such
as over-current, over-voltage, earth-fault, etc. to ensure safe
and reliable operation. However, as the tasks performed by
these machine grew increasingly complex; improvements
were also sought in the field of fault diagnosis. It has now
become very important to diagnose faults at their very
inception; as unscheduled machine downtime can upset
deadlines and cause heavy financial losses.
The major faults of electrical machines can broadly be
classified as the following [l]:
a) Stator faults resulting in the opening or shorting of one
or more of a stator phase winding,
b) Abnormal connection of the stator windings,

0-7803-5589-X/99/$10.00Q 1999 IEEE

197

c) Broken rotor bar or cracked rotor end-rings,


d) Static and /or dynamic air-gap irregularities,
e) Bent shaft (akin to dynamic eccentricity) which can
result in a rub between the rotor and stator, causing
serious damage to stator core and windings,
Shorted
rotor field winding ,and
f)
g) Bearing and gearbox failures.
These faults produce one or more of the symptoms as
given below:
a) Unbalanced air-gap voltages and line currents,
b) Increased torque pulsations,
c) Decreased average torque,
d) Increased losses and reduction in efficiency, and
e) Excessive heating.
The diagnostic methods to identify the above faults may
involve several different types of fields of science and
technology. They can be described as L1-21:
a) Electromagnetic field monitoring, search coils, coils
wound around motor shafts (axial flux related detection),
b) Temperature measurements,
c) Infrared recognition,
d) Radio frequency (RF) emissionsmonitoring,
e) Noise and vibration monitoring,
f) Chemical analysis,
g) Acoustic noise measurements,
h) Motor current signature analysis (MCSA),
i) Model, artificial intelligence and neural network based
techniques.
Of the above types of faults i) bearing, ii) the stator or
armature faults, iii) the broken rotor bar and end ring faults of
induction machines and iv) the eccentricity related faults are
the most prevalent ones and thus demand special attention.
Thus, these faults and their diagnosis techniques will be
discussed briefly in the next section. A brief introduction to
fault detection using artificial intelligence. (AI) techniques has
also been included.

U. VARIOUS -PIS

OF FAULTS
AND THEIR
DETECTION

TECHNIQUES
A. Bearing faults
The majority of the electrical machines use ball or rolling
element bearings. Each of these bearings consists of two

rings, one inner and the other outer. A set of balls or rolling
elements placed in raceways rotate inside these rings [Z].
Even under normal operating conditions with balanced load
and good alignment, fatigue failures may take place. These
faults may lead to increased vibration and noise levels.
Flaking M spalling of bearings might occur when fatigue
causes small pieces to break loose from the bearing.
Other than the normal internal operating stresses, caused
by vibration, inherent eccentricity, and bearing currents [39]
due to solid state drives, bearings can spoiled by many other
external causes such as
a) Contamination and corrosion caused by pitting and
sanding action of hard and abrasive minute particles or
corrosive action of water, acid etc.
b) Improper lubrication; which includes both over and under
lubrication causing heating and abrasion.
c) Improper Installation of bearing. By improperly forcing
the bearing onto the shaft or in the housing (due to
misalignment) indentations are formed in the raceways
(brinelling).
Though almost 4040% of all motor failures is bearing
related, very little has been reported in literature regarding
bearing related fault detection. Bearing faults might manifest
themselves as rotor asymmetry faults [Z], which are usually
covered under the category of eccentricity related faults.
Otherwise, the ball bearing related defect? can be categorized
as [I] outer bearing race defect, inner bearing race defect, ball
defect and train defect and the vibration frequencies to detect
these faults are given by,
f,[HzI = ( N I 2 )f,U- bd COS(P11d,I
for an outer bearing race defect
f,[HzI= ( N / 2 ) f , [ I + bdCOSV
) d, 1
for an inner bearing race defect

Yazici et.al. [5] have reported of an adaptive, statistical


time ffequency method for detection of bearing faults.
Experiments were conducted on defective bearings with
scratches on the outer races and bearing balls and cage
defects. It has been claimed that all defective measurements
were correctly classified as defective. However, the detection
procedure required extensive training for feature extraction.

B. Stator or armature faults

These faults are usually related to insulation failure. In


common parlance they are generally known as phase-toground or phaseto-phase faults. It is believed that these faults
start as undetected turn-to-turn faults which finally grow and
culminate into major ones 161. Almost 30-40 % of all
reported induction motor failures falls in this category [6].
Armature or stator insulation can fail due to several
reasons. Primary among these are [Z]
a) High stator core or winding temperatures.
b) Slack core lamination, slot wedges and joints.
c) Loose bracing for end winding.
d) Contamination due to oil, moisture and dirt.
e) Short circuit or starting stresses.
0 Electrical discharges.
g) Leakage in cooling systems.
There are a number of techniques to detect these faults.
Penman et. al. [7] were able to detect turn to turn faults by
analyzing the axial flux component of the machine using a
large coil wound concenuically around the shaft of the
machine. Even the fault position could be detected by
mounting four coils symmetrically in the four quadrants of
the motor at a radius of about half the distance from the shaft
to the stator endwinding. The frequency components to
detect in the axial flux component is given by,
f,[ffzI= dpf, Ib,(l-[b, COS(P) /d,IZ)
(3)
( k f n(1- s) / p ) f
for a ball defect
where p is the number of pole pairs, f i s the mains
fJffzI= (f,/2)f,[1-bd c W P ) I d p ]
frequency, k = 1,3 and n = 1,2,3,...,,(2p-l) and s is the slip.
for a train defect
(1)
Toliyat and Lip0 [8] have shown through both modeling
where f, is the rotational frequency, N is the number of and experimentation that these faults result in asymmetry in
balls, bd and dparethe ball diameter and ball pitch diameter the machine impedance causing the machine to draw
unbalance phase currents. This is the result of negative
respectively, and p is the contact angle of the ball (with the sequence currents flowing in the line as also have been shown
races).
in [ 9 ] . However, negative sequence cunents can also be
Schoen et. al. [3] have shown that these vibration caused by voltage unbalance, machine saturation etc. Kliman
ffequencies reflect themselves in the cunent spectrum as
et.al. [6] model these unbalances which also includes
instrument asymmetries. It is reported that with these
(2) modifications it is possible even to detect a one turn bolted
f b n g = If,+ m.f
fault out of a total 648 turns. Statistical process control (SPC)
where m = 1,2,3,... and f, is one of the characteristic techniques have also been applied to detect stator faults 1301.
vibration frequencies. However, the experimental results
were presented for rather extensive bearing damages (such as C. Broken rotor bar and end ring faults
hole in the outer race of the bearing; brinelling induced by a
Unlike stator design, cage rotor design and manufacturing
vibration table). The implementation of an unsupervised online detection of these faults using artificial neural networks has undergone little change over the years. As aresult rotor
has also been described in [41.

198

failures now account for around 5.10% of total induction


motor failures (Bonnett [lo], Kliman [6, 111).
Cage rotors are of two types: cast and fabricated.
Previously, cast rotors were only used in small machines.
However, with the advent of cast ducted rotors; casting
technology can be used even for the rotors of machines in the
range of 3000 kW. Fabricated rotors are generally found in
larger or special application machines. Cast rotors though
more rugged than the fabricated type, can almost never be
repaired once faults like cracked or broken rotor bars develop
in them.
The reasons for rotor bar and end ring breakage are
several. They can be caused by
a) Thermal stresses due to thermal overload and unbalance,
hot spots or excessive losses, sparkjng (mainly fabricated

external search coil wound around the shaft of a machine.


The frequency components are still given by (6)with k =1,2,3.
Modeling of rotor bar and end ring faults have been described
in [8]. Broken bar detection using state and parameter
estimation techniques have also been reported [26].However
the current spectrum and the parameter estimation approach
have been compared and the former has been found more
efficient [U].
As suggested in 131-321,presence of interbar currents in
n

rotors),
b) Magnetic stresses caused by electromagnetic forces,
unbalanced magnetic pull, electromagnetic noise and
vibration,
c) Residual stresses due to manufacturing problems,
d) Dynamic stresses arising from shaft torques, centrifugal
forces and cyclic stresses,
e) Environmental stresses caused by for example
contamination and abrasion of rotor material due to
chemicals or moisture,
f) Mechanical saesse due to loose laminations, fatigued
parts, bearing failure etc.
Kliman [11], Thomson [12], Filippetti [13], Elkasabgy
[14] used spectrum analysis of machine line current (MCSA)
to detect broken bar faults. They investigate the sideband
components,f,, around the fundamental for detecting broken
bar faults.
(4)
fb = (1 f 2 s ) f
While the lower sideband is specifically due to broken bar,
the upper sideband is due to consequent speed oscillation. In
fact, [13] shows that broken bars actually give rise to a
sequence of such sidebands given by
fb = (1kZks)f , k=l,2, 3, ...
(5)
The motor-load inertia also affects the magnitude of these
sidebands. Other spectral components that can be observed in
the stator line current is given by 1111 and Gaydon [15]

-40.

199

300

-5.8

-f

6.2

6.4

lo

Frequency (Hz)

Time (Secs.)
182.6r
1
8

4w

350

Frequency (Hz)

Frequency (Hz)

6
8
O
.u
0,

;r 182.4
0

;
182.2
a

L-lw

v)

186

-150

6.1
6.2
Time (Secs.)

10

Frequency (Hz)

Pi@ Simulated plots of normalized line cunenl specIra around


fundamental and 5 and 1lime harmonic (lop IOW), torque and its specua
(middle IOW), speed and ils spectra @atom row) with two bars paaially
broken. Slip = 0.033.
10
- 5

8
E

=z -5

1181.6

-10

where, f b= detectable broken bar frequencies; Up = 1,3,5..


Elkasabgy 1141has also shown that broken bar faults can
also be detected by time and frequency domain analysis of
induced voltages in search coils placed internally around
stator tooth tip and yoke and externally on motor frame. The
frequency components are given by (6) with k = l . Torque and
speed signals also contain 2s and 4$ frequency
components with broken rotor bars [13-14].Following the
works of Penman [16], detection of these faults are also
possible by frequency domain analysis of shaft flux or more
generally axial leakage flux which is monitored by using an

-OW.

1W

50

6.1

6.2

v1

-50
6

6.1

6.2

Time(Secs.)

-3

-50

U1

-1W

-lW
0

50

-150

1W
0
Frequency (Hz)

10

Fig 2. Simulated plots of line current and speed (top row) and their
normalized spectra (bottom row) for the IWO end ngs panially broken. Slip
= 0.036.

uninsulated rotor cages, where the contact between the rotor


core and the bars are good, might make broken bar detection
difficult.
Fig.1 and Fig.:! show simulated current, speed and torque
waveforms and their related spectra with two partially broken
bars and two partially broken end ring faults for a 3ph, 3hp,

-201
-40
0

-60
-80
-100

40

30

50

60

80

70

90
Frequency (Hz)

-2

-20

Pig 5. Experimental plots of normalized speed spectra of healthy machine


(top) and with four bars broken @mom). Slip = 0.033.

II

-40
-401

-60

60 Hz, 4 pole, skewed 44 rotor bar induction motor. The


current components given by (4-6) and the 2sf and

-80
-100

40

30

50

60

70

80

90

4sf speed related components

-20

Q
0
2

-40
-60

-80
-100
30

40

50
60
70
80
90
Frequency (Hz)
Fig 3. Expcrimental plots of n o r d i z e d line cul~entspectra of healthy
machine (top) and with two (middle) and four bars partially broken (bottom).
Slip = 0.033. These and the four subsequent plots are obtained from a
machine that is similar to the one simulated. Faults were introduced by
drilling the bar*.

350
-40

0 -60

4w

450

n
(0
a -60
-1w

3w

350
4w
Frequency (Hz)

450

Fig 4. Experimental plots of normalized line arrent spectra of healthy


machine (top) and with four bars broken @ottom) around the 5' and 7' time
harmonics. Slip = 0.033.

200

can be clearly seen in the

plots.
However, in practice, the current sidebands around
fundamental may exist even when the machine is healthy, as
can be seen in Fig 3. This could be due to uneven rotor bar
resistance because of the die casting process, rotor
asymmetry etc. Also components given by (5) may not show
any marked change (Fig. 4). Hence, at least for small motors,
it may be worthwhile to c o n f m the presence of broken bars
through the speed spectra Ojig.5).
D. Eccentriciry related faults

Machine eccentricity is the condition of unequal air-gap


that exists between the stator and rotor ([l], Cameron 1171).
When eccentricity becomes large, the resulting unbalanced
radial forces (also known as unbalanced magnetic pull or
UMP) can cause stator to rotor rub, and this can result in the
damage of the stator and rotor. There are two types of air-gap
eccentricity: the static air-gap eccentricity and the dynamic
air gap eccentricity. In the case of the static air-gap
eccentricity, the position of the minimal radial air-gap length
is fixed in space. Static eccentricity may be caused by the
ovality of the stator core or by the incorrect positioning of the
rotor or stator at the commissioning stage. If thz rotor-shaft
assembly is sufficiently stiff, the level of static eccentricity
does not change.
In case of dynamic eccentricity, the center of the rotor is
not at the center of the rotation and the position of minimum
air-gap rotates with the rotor. This misalignment may be
caused due to several factors such as a bent rotor shaft,
bearing wear or misalignment, mechanical resonance at
critical speed, etc. Dynamic eccentricity in a new machine is
controlled by the total indicated reading (TIR) or "run-out" of

the rotor (Thomson, [191). An air-gap eccentricity of up to


10% is permissible. However, manufacturers normally keep
the total eccentricity level even lower to minimize UMP and
to reduce vibration and noise ,
In reality, both static and dynamic eccentricities tend to coexist. An inherent level of static eccentricity exists even in
newly manufactured machines due to manufacturing and
assembly method, as has been reported by Dorrell [18]. This
causes a steady UMP in one direction. With usage, this may
lead to bent rotor shaft, bearing wear and tear etc. This might
result in some degree of dynamic eccentricity. Unless
detected early, these effects may snowball into stator to rotor
hub causing a major breakdown of themachine [19].
The presence of static and dynamic eccentricity can be
detected using MCSA 11,171. The equation describing the
frequency components of interest is
(7)
where nd = 0 in case of static eccentricity, and nd = 1,2,3.,.
in case of dynamic eccentricity ( n d is hown as eccentricity
order), f is the fundamental supply frequency, R is the
number of rotor slots, s is the slip, p is the number of pole
pairs, k is any integer, and Y is the order of the stator time
harmouics that are present in the power supply driving the
motor ( Y = t 1, + 3, ? 5, etc.). In case one of these harmonics
is a multiple of three, it may not exist theoretically in the line
current of a balanced three phase machine. However it has
been shown by Nandi [20], Ferrah [21] that only a particular
combination of machine pole pairs and rotor slot number will
give rise to significant only static or only dynamic
eccentricity related components. This relationship for a 3ph,
integral slot, 60 desee phase belt machine is given by:
R = 2 p [ 3 ( m f q ) k r ] i .k
(8)
where m f q = O,1,2,3,... and r = Oor 1, k = 1.
Equation (8) assumes only the fundamental eccentricity
component in the permeance or inverse air-gap function
[21,43].
It may also be possible to detect these components in other
machines, for example for those matching (8) with k = 2 .
However, some of these components are noticeable only
under light load conditions.
Simulated results with a 4 pole, skewed, 43 rotor slot
machine, which conforms to (8) with k = 1 are given in Fig.
6 . Similar results with a 4 pole, skewed, 42 rotor slot machine
( k = 2 ) are given in Fig.7. The effects of eccentricity on
frequency components given by (7) seem to be much less
pronounced for this machine.
It has also been ascertained that machines generating
principal slot harmonics (PSH) will not give rise to these
components with only static or only dynamic eccentricity.
The pole pairs and rotor slot numbers for these machines
(3ph, integral slot, 60 degree phase belt) are related by:

R =2p[3(mi.q)+r]

(9)

where mlt q = O,1,2,3,... and r = 0 or 1


However, if both static and dynamic eccentricities exist
together, low frequency components near the fundamental
[18,22] given by

=If

f,
+kf,I ,
k = 1 , 2 , 3 ...
(10)
can also be detected for all machines (Fig.8). These low
frequency components also give rise to high frequency
components as described by (7). However, these components
are strong only for machines (Fig.9) whose pole pairs and
rotor slot numbers are given by (8) ( k = 1 ) and (9). For
machines described by (8) with k = 2 they are rather weak
(Fig.10).
Modeling based approaches to detect eccentricity related
components in line current have been described in [20,22].
The simulation results obtained through the models are also
well supported by permeance analysis and experimental
results. Fig.11 shows the experimental results for a similar,
skewed, 4 pole machine with R=44 and a nominal 38.46%
static and inherent dynamic eccentricity. The low frequency
sidebands that are present even under healthy condition did
not show appreciable change once eccentricity was
introduced. However, the high frequency components showed
moderate (around 5dB) increase.
Vibration signals can also be monitored to detect
eccentricity related faults. The high frequency vibration
components for static or dynamic eccentricity are given by
[17] using an equation similar to (7) (only the values of nd
and V are different). In case of mixed eccentricity, the low
frequency stator vibration components are given by,
f" =2f +f,
(11)
Time stepping finite element methods have been employed
recently to compare simulated results with experimentally
obtained static eccentricity components in line currents [19].
Static eccentricity has also been modeled using Winding
Function Approach [36].
Other approaches, such as monitoring the stator voltage
and current park's vector (Cardoso [Z3]) to detect eccentricity
in induction motor, can also be found in literature. Toliyat
and Al-Nanim [24] have provided simulation and
experimental results for synchronous machines with dynamic
eccentricity related faults.
III. ARTIFICIAL.
INTELLIGENCE
(AI) BASEDMACHINE
CONDITION
MONITORING
ANDFAULTDIAGNOSIS
In the recent past artificial neural networks (ANN), fuzzy
or neuro-fuzzy systems are being extensively used for speed,
torque estimation, solid state drive control of both dc and ac
machines [41]. However, they are particularly suited for ac
machines' applications where the relationships between
motor current and speed are non-linear.
These AI techniques are now being increasingly used for
condition monitoring and fault detection of machines
[4,13,41]. A neural net based fault diagnosis system utilizing

201

the stator current spectra is described in Fig 12. The


preprocessor extracts the fiequency components of the
sampled current data. Using the rule based frequency filters
these frequency components are classified into four
categories with decreasing level of importance. Based on
these rules, a neural network, which has been trained for all
possible operating condition oi the machine, is used to
classify the incoming data. A spectral signature that falls
outside the trained clusters are marked as potential motor
fault. In order to prevent false diagnosis, the postprocessor
sends an alarm only when fault signatures are observed
persistently. This function is performed by maintaining a time
history of the motor being monitored. Such a scheme has
been successfully implemented [4] to diagnose bearing and.
unbalanced rotor faults of induction motors.

-80

v)

- 1 ~
-120
1200
-40 I

1250

1300

-50
U)

0.

-lW

50

1W

1400

-50

'

,
1250

1300

-120p
1200

ft21r

1450

DEC

-80

150

fundamental
4 f+f'

1350

.60t

-601

Similarly, fuzzy logic based systems has been used [41] to


classify broken bar related faults by categorizing the two
sideband components (4) around the fundamental component
of the induction motor line cunent by a set of nine rules.
Denoting the sidebands as A1 and A2 which are the two
inputs of the system and n, the number of broken bars as the
output of the system, an example of t h s e rules is " If AI is

,.I

1350

, , ,..

-100

50

1400

100

150

1450
Frequency (Hz)

Frequency (Hz)
Fig.6 Simulated normalized plots of the line curmot spctra of a 3ph, 3hp, 60
Hz induction motor with 38.46% Static (top) and 20% dynamic eccentricity
(bottom) with Ips;R=43. Slip =0.029.

$.^

Fig.8 Simulated, normalized line current spedra of 3ph, 3hp, 60 Hz, skewed,
4 pole induction motors with different rotor slats and identical mined
eccentricity (SE =38.46%, DE=20%) machine around fundamental. From
top to bottomR=44,43,42. Slip=O.O29.

SEC

L -lW
-120
1150

-60

xc .1W

1200

DECI

. I

1300

1250

1150

1250

1300

1350

1400

-120t1
1200

1250

1300

1350

1400

,,

1450

DEC2

-120

1100

1200

1200

1250

1300

1350

1400

Frequency (Hz)

1450

Frequency (Hz)

Rg.7 Simulated normalized plots of the line current spema of a 3ph, 3hp, 60
Hz induction mota with 38.46% &tic with Slip= 0.029 (top) and 40%
dynamic eccentricity and Slip =0.00467 (bottom) with 2 e ; R=42. The
other static eccentricity component is suppressed due to loading effects.

202

Pig9 Simulated, normalized high frequency line current spectra of 3ph, 3hp,
60 Hz, 4 pole induction motors with different rotor slots and identical mixed
eccenllicity (SE ~38.46%DE=20%) machine. From top to bottom R 4 4 ,
43. Slip=0.029.

-n
Ln

small and A2 is large n equals one broken b d . The fuzzy


logic system considered is the Mamdani type. The fuzzy
inference is performed by using the fuzzy implication minmax methods and the centroid defuzzification technique is
used. The membership functions for AI and A2 are small,
medium and large. Other examples of motor fault detection
using neural networks and fuzzy logic techniques can be
found in [42].

1[0.5(R.1)(1-~)-1]

SEC1

I
.1W
.110

-120
1100

1120

1140

1160

1180

fI0.5(R+l)(l-s)+ll

.I20
1200

1250

1300

1350

1400

1200

IV. CONCLUSIONS

1450

Frequency (Hz)
Fig.10 Simulated, normlimed high frequency line cumen1 spectra of 3ph,
3hp, 60 Hz, 4 pole induction motors with 42 rotor slots and mixed
eccentricity (SE =38.46%,DII;ZO%) machine. Sli0;0.029.
0

-401

Psn

m^
p -50
n

A brief review of bearing, stator, rotor and eccentricity


related faults and their diagnosis has been presented in this
paper. It is clear kom various literatures that non-invasive
motor current signature analysis (MCSA) is by far the most
preferred technique to diagnose fault. However, theoretical
analysis and modeling of machine faults are indeed necessary
to distinguish the relevant frequency components tiom the
others that may be present due to time harmonics, machine
saturation, etc. Other techniques for fault detection such as
axial flux based measurements, vibration analysis, etc. have
also been discussed. A section on automated fault detection
has also been included.

-lW

0
0

50

1W

150

-40

8 -50
8n

ACKNOWLEDGMENT
1300 1350 1400 1450

Psn

This material is based in part upon work supported by the


Texas Advanced Research Program under Grant No. 95-PO83
and by the Department of Energy under Grant No. DE-FG0798ID13641.

-60

-80

-100

-1W

-120

50

1W

150

REFERENCES

1300 1350 1400 1450

P. Vas, Pormeler Exlimfion, Condition Monitoring,and Diagnosis


oJElectrical Machiner,Clilrendron Press, Oxford, 1993.
121 G . B. Kliman and 1. Stein Indimion molor fault delection via
p m ~ i v e cuiienl monitoring, International Conference in Electrical
Machines, Cambridge, M k pp.13-17, August 1990.
R. R. Schoen, T. G. Habetler, F. Kaman, R. G . Banhcld, Motor
131
bearing damage detection using stator current monitoring, IEEE
Tranr.Ind.Applnr., ~01.31,no. 6, pp 1274-79, Nov-Dec 1995.
141 R. R. Schoen, B. K.Lin, T. G. Habetler, J. H. Schlag, S . Farag, An
unsupervised on-line rynem for induction motor fault detection using
stator current monitoring, IEEE Trans. lnd. Applnr., ~01.31,no. 6, pp
1280-86, Nov-Dec 1995.
B. Yazid, G . B. Kliman, W. 1. Premerlani, R. A. Koegl, G . B.
(51
Robinson and A. Abdel-Malek,
An adaptive. on-line, statistical
method far bearing fault detection using stator anent, IEEE-US
Annual Meeting conference, pp 213- 220, New Orleans,LA, 1997.
161 G. B. Kliman, W. 1. Premerlani, R. A. Koegl and D.Hoeweler, A
new approach to on-line fault detection in ac motors, IEEE-aS
Annual Mceling ConJemce, pp.687-693. San Diego, CA, 1996.
171 1. Penman, H. G . Sedding, B . A. Lloyd, W. T. Fink. Detection and
location of intenurn *hat circuits in the stator windings of operating
motors, IEEE Trans. Energy Conv., vol.9, no.4, Dec 1994.
IS1 K. A. Tdiyat and T.A. Lipo, Tranrient analysis of cage inducUon
machines under stator, rotor bar and end ring faults, IEEE Tram.
Energy Conv.. vol.10, no.2, June 1995.
191 S. Williamson and P. Mirzoian, Analysis of cage induction motor
with Etator winding faults. IEEE-PES. Summer Meeing, July 1984.
I11

Preprocessor
FFT & Averaging

Postprocessor
Time History
and Alarms

Rule-B ased
Frequency
Filters

Clutsering
Neural Network
Algorithm

203

I101 A.H. Bonnett and G. C. Soukup, "Rotor failures in Squirrel Cage


induction motod'. IEEE T r m . Ind. Applns.., vol. IA-22, 110.6, pp.
1165.1173, Nov./Dec. 1986.
[ I l l G.B. K h a n R A . Koegl, J. Stein, R. D. Eodicott. M. W. Madden
"Noninvasive detection of broken rotor bars in operating induction
motors", IEEE Tram. Energy Cony. vol. EC-3, no.4, pp.873-879, Dec.
1988.
1121 W.T. Thornon, and I. D. Stewart, "On-line current monitoring for
fault diagnosis in inverter fed induction motord', IEE Third
international conference on power electronics and drives, London,
pp.432435, 1988.
[131 F.Filippctti, G. Franceschini, C. Tassoni, P. Vas, "A1 techniques in
induction machines diagnosis including the speed ripple effect",
IEEE-IAS Annual meeting conference, San Diego, QCI, 655-662. Oct 6 IO, 1996.
I141 N.M. Elkasabev. A. R. Eastham. G. E. Dawson. "Detection of broken

defects by speed fluctuation measurements", Eleclric test and


measuring instrumentation Tesrmer '79 conference paper*, pp 5.8,
1979.
I161 I. Penman. M. N. Dey, A. I. Tail, W. E. Bryan, '' Condition monitoring
of electrical drives". IEE Proceedinns. on, 142.148. vol. 133. ot. B..
110.3, May 1986.
1171 J. R. Cameron, W.T. Thonuon, and A. B. Daw, "Vibration and current
monitoring for detecting airgap eccentricity in large induction
motors," IEE Proceedings. pp. 155.163, val ,133, pt. B, no.3, May
1986.
1181 D. G. Dorrell, W. T. Thomson and S . Roach, "Analysis of airgap flux,
current, vibration signals as a function of the combination of static and
dynamic airgap eccentricity in 3-phasz induction motors". IEEE
l'ruiu. Ind.App1nu.n.. vol. 33, No.1, pp. 24-34, 1997.
1191 A. Barbour and W.T. Thomson, "Finite element study of rotor dot
designs with respect l o ament monitoring for detecting static airgap
eccentricity in squirrel-cage induction motor," IEEE-IAS annual
meeting conference recordings, pp. 112-119, New Orleans, Louisiana,
06.5-8, 1997.
[ZO] S . Nandi and H. A. Toliyat, "Detection of Rotor Slot and Other
Eccatricity Related harmonics In a three Phase Indoclion Motor With
Different Rotor Cages." IEEE-PEDES'%? Conference Proceedings,
Perth, Australia, Nov.30-Dec.), 1998.
lzl] A. Fernah, P.J. Hogben-Liang, K. J. Bradley, G.M. Asher, M. S.
Woolfson, "The effect of rotor design of sensorless speed eStimation
using rotor slot harmonics identified by adaptive digital filtering using
the maximum likelihood approach," IEEE-/AS annual meeting
conference recordings, pp. 128-135, New Orleans, Louisiana, Oct.58, 1997.
1221 S. Nandi, RajMohan Bharadwaj, H.A. Toliyat, A.G. Parlos,
"Performance analysis of a three phase induction motor under
incipient mixed eccentricity condition", IEEE PEDJ3.98 conference
proceedings, Perth, Australia, Nov.30-Dec.3, 1998.
I231 A. I . M. Cardoso and E. S . Saraiva, " Computer-aided detection of
aireaD eccentricitv in ooeratine three-ohase induction motors bv
PaFYi Y ~ C ~ Oappr&ch",
I
iEEE fmm. I&
Applm.,
. pp. 897.901, voi.
29, "0.5, Sept./Oct. 1993.
I241 H.A. Toliyat, N. A. AI-Nuaim, "A novel method for modeling
dynamio air-gap eccentricity in synchronous machines based on
modified winding function theory," presented at IEEE-PES S u m e r
Mcerin& July, 1991.
1251 P. I. Tavner and J. Penman, Condition Moniloring of Elecfrical
Muchines, Research Studies Press,Letchwonh, England, 1987.
1261 K. R. Cho, J. H. L a g , and S.D. Umans, " Detection of broken rotor
bars in induction motors using state and pnramneter &mation," IEEE
Trans. Industry Applns, pp.702-709, Vol.28, No. 3 MayNune, 1992.
(271 F. Filippetti, G. Franceschini, C. Tassoni, P. Vas, " Broken bar
detection in induction machines: comparison between current
spectrum approach and paramet,% estimation appr~ach",Proceedings
of lhe IEEE -1AS Annual Meeting Conference. pp. 95-102, Vol. 1,
Denver, Colorado, USA, 1994.

I.

..

[ Z R ] D.I. T. Siyambalapitiya, P.G. MeLare& '' Reliability improvement and


economic benefiu of on-line monitoring systems for large induction
machines". IEEE Tram. Indumy Applns, pp.1018-1025. Vo1.26, No.
6 Nov./Dec., 1990.
1291 I. L. Kohler, J. Sottile, F.C. T N ~ .'"Alternatives for assessing the
electrical integrity of indunion motors", IEEE Trans. Industry Applns,
pp. 1109-1117,Vo1.28,No. 5 Sept./Oct., 1992.
1301 C.J. Dister and R. Schiferl, '* Using temperature, voltage, andlor speed
measurements to improve trending of induction motor rms currents in
QIOCCSS control and diagnostics", '
0 Proceeding8 of the IEEE -1AS
Annual Meeting Conference,pp. 312-318, Vol. 1, St. Louis, Missouri,
USA, 1994.
[31] I. Kermnbaum and C. F. Lmdy, " The existence of large inter-bar
currents in three phase squirrel cage motors with rotor-bar andlor endring faults". E E E T~onsactionson Power Appamru* and Systems,
00.1854-1862. Vol. PAS-103. No.1. July 1984..

Mar. 1994.
1331 I. Penman, C.M. Yin, "Feasibility of using unsupervised learning,
artificial neural networks for the condition monitoring of electrical
machines", IEE Proceedings on Elect,. Power Applns., pp. 3 17-322,
Vol. 141, No.6, Nov.1994.
1341 F. Filipptti, G. Franceschini, C. Tassoni, Neural networks aided online diagnostics of induction mMor faults", Proceedings ofthe IEEE /AS Annual Meeting Conference, pp. 316-323, Vol. 1, Toronto,
Canada 1993.
P. L. Alger, The Nature of Induction Machine.<,Gordon and Breach,
New York, 1965.
H. A. Toliyat, M. S. Arefeen, A. G. Parlos, "A method for dynamic
simulation of *-gap eccentricity in induction machines," IEEE Trans.
Ind.Applm.,pp. 910-918, Vol. 32.No.4, July/Aug., 1996.
1371 S.Nandi. H.A.Taliyat, "Performance analysis of a lhree phase
induetion motor under abnormal operating conditions", IEEESPDEMPED. Franee, 1997.
I381 G. Kron, Equivalent circuit3 of electric machinery, John Wiley &
Sons. Inc.. New Yak, 1951.
1391 S . Chen, T.A. L i p . '"Bearing m e n t s and shaft voltages of an
induction motor under hard- and soft-switching inverter excitation",
IEEE Trans. Ind. Applm., pp. 1042-1048, Vol. 34, No. 5, Sept/O~t.,
1998.
1401 M. M. LiwschitZ " Field harmonics in induction moto~s",Trans. Am.
hf.
E k .Engg.,pp. 797-803,VOl. 61, Nov., 1942.
1411 P. Vas, Arrificinl -infelligence-based electrical machines and drives:
applications of fuzzy, neural, fay-neural and genetic algorithm
based Iedmiques, Oxford University Press, New York, 1999.
I421 M. Y. Chow, Methodologies of using neural network and fuzzy logic
technologies for m l o r incipient fault detection, World Scientific
Publishing Co. Pte. Ltd. Singapore, 1997.
143) B. Heller and V. Hamata, Harmonicfielde n e m in induction machine.
New York, 1977.

204