2020 IEEE International Conference on Computing, Power and Communication Technologies (GUCON)

Galgotias University, Greater Noida, UP, India. Oct 2-4, 2020

An Improved Matrix Pencil Method based Bearing

Fault detection in Three Phase Induction Motor
K. C. Deekshit Kompella G. Venu Madhav
EEE Department EEE Department
Sreenidhi Institute of Science & Technology ANURAG Group of Institutions
Hyderabad, India Hyderabad, India
[email protected] [email protected]

Abstract—This paper presents an improved fault detection presented using Wigner Ville Distribution (WVD) and Short-
criteria for bearing faults in a three phase induction motor time Fourier transform (STFT) in [7]-[8]. But drawbacks asso-
using current signature analysis. The motor current is extracted ciated with this techniques like more mathematical calculation
under various conditions of the machine and used for analysis
in frequency domain. To overcome the resolution problems and spectral leakage due to fixed window dimensions will
associated with current signature analysis, the signal is pre- leads to the development of variable sized window based
processed using advanced signal processing tools. High resolution spectral analysis (Wavelet Transforms) and has been presented
techniques like matrix pencil method and wavelet de-noising in [9]-[10]. This has shown very good performance in many
are implemented to estimate the fault severity. Conventional situations but have few problems due to severe noise in
pencil method have not shown good indication for fault due
to domination of pre-fault components like fundamental and the signal. In [11], the fault detection using SWT has been
harmonics. The Wiener filter cancellation is used to remove presented and have few problems especially for incipient
these pre-fault components and the fault indexing parameter is stage faults. For faults at severe stage can be estimated using
calculated to estimate the fault existence and its development. proposed technique in [11] with high reliability. But, due to
The proposed fault detection criteria is tested using experimental the severe noise presented in the extracted current signal, the
verification on 1.5kW induction motor.
Index Terms—Bearing Fault, current signature analysis, ma- incipient stage faults have not be identified using SWT based
trix pencil method, pre-fault components, Wiener filter cancella- diagnosis.
tion Moreover, the domination of pre-fault (Before fault exis-
tence) components is another major issue in current based
I. I NTRODUCTION monitoring. This problem is addressed in [12]-[15] using
Notch filter, frequency spectral subtraction, Park’s vector
Huge usage of induction motors in all ranges of industries
transformation and Wiener filter based calculation. But they
will leads to the requirement of more concentration to avoid
have their own draw backs like, the Notch filter will remove
sudden damage of the machine. Due to the strong relation be-
the side bands of the pre-fault components and may remove
tween health condition of machine and production of industry,
the nearest fault component. In FSS method, the subtraction
continuous monitoring is required over periodic monitoring.
is not accurate if there is a slight variation in fundamental
For continuous monitoring of the machine, cost effective,
component and may lead to wrong diagnosis of the machine.
non-intrusive and easy technique should be required. Among
In the similar way the Park’s vector transformation requires
various monitoring techniques like vibration monitoring, noise
the three phase currents which intern requires more number
monitoring, chemical monitoring etc, current monitoring will
of current sensors. Where as the Wiener filter has shown
meet all these requirements and have many advantages like
good performance in [15]-[16], the pre-fault components are
requires less man power, easy to access and does not requires
not completely removed as compared to fault component. As
any additional costly sensors. Therefore, many researchers [1]-
a result, the fault components are not elevated sufficiently.
[17] have concentrated on current based monitoring of bearing
This requires another spectral technique like FFT, wavelet
fault detection. Conventional FFT based current monitoring
transforms etc along with Wiener filter as presented in [17].
has few problems like less resolution, spectral leakage, unable
Further, the fault indexing parameters like standard deviation,
to provide time-frequency relation etc[1]-[3]. To overcome
Energy Entropy and Simple square integrals are required to
the drawbacks associated with conventional FFT many ad-
estimate the fault severity.
vanced spectral analysis techniques have been presented in [4]-
Therefore in the present work, the bearing fault diagnosis
[11]. The resolution problems are addressed in [4]-[6] using
is carried out using proper combination of suitable pre-fault
ZFFT(ZOOM), MUSIC algorithm and Matrix Pencil method.
component cancellation and spectral technique along with
The time-frequency relation based spectrum analysis has been
feature extraction parameter as fault indexing. The Wiener
JNTUH under TEQIP-III grants with sanction order No. JNTUH/TEQIP- filter is used to cancel the pre-fault components in real time
III/CRS/2019/EEE/04 fashion and the Matrix Pencil method is used as spectral

978-1-7281-5070-3/20/$31.00 ©2020 IEEE 51

Authorized licensed use limited to: Rajamangala Univ of Technology Srivijaya provided by UniNet. Downloaded on August 31,2023 at 06:48:58 UTC from IEEE Xplore. Restrictions apply.
 component cancellation. Wavelet de-noising is carried out to The mathematical expressions for decomposing stator current
cancel the higher order noise developed during extraction of is presented in the following [18].
current signals from the machine as a pre-processing tool. The 
fault severity is estimated using feature extraction parameter Aj (m) = L(n − 2m)i(n) (1)
of Energy of the processed signal. n

II. FAULT D ETECTION C RITERIA Dj (m) = H(n − 2m)i(n) (2)
The proposed fault detection criteria is carried out using n
four stages of signal processing as shown in “Fig. 6”. After Where L is low pass filter coefficients, H is the high pass
the current signal extracted from the machine, the signal is filter coefficients of j th order decomposition, and i(n) is the
processed in four stages before making the decision about current signal.
the fault to improve the reliability of the proposed system.
In the first stage, the signal is de-noised using DWT with four B. Stage-2 (Wiener Filter Cancellation)
levels of decomposition. After de-noising the current signal, The de-noised stator current is processed for second stage
the healthy components of the stator current is removed using processing using Wiener filter to remove dominated pre-
Wiener filter to reduce the domination of fundamental com- fault components. As the machine at under normal operating
ponent and its harmonics. In stage-3, Matrix pencil method condition, the stator current spectrum consists of fundamental
(MPD) is employed to calculate the spectral components in component along with its harmonics and a wide spread noise
the current signal after remove of dominated components. In due to other sources namely EMI and sensor devices. These
fourth stage, the feature extraction parameter Energy of the components will have large magnitudes and dominates the
Pencil components of the stator current is evaluated to estimate fault components especially at premature stage. Which leads
the fault. Finally, the fault existence is announced using fault to imperfect prediction of faults and may effect the economy
indexing criteria. The mathematical analysis of these steps are of industry. As a consequence, it is needed to elevate the fault
discussed in the following subsections. component by abolishing the pre-fault components. To furnish
the requirement, Wiener filter based cancellation is adopted
from [15] to estimate and eliminate the healthy components.
Start
Extraction of stator current from From the Fig.2 it is observed that, the filter is said to
three phase induction motor
perfect if the predicted components of stator current are
same as actual components in the stator current after de-
DWT based Wavelet De-Noising
Stage-1 to remove noise developed noising. But practically the predicted components are differ
while extraction of signal from the actual components due to problems associated with
design of filter. To achieve best performance of the filter,
Stage-2
Wiener Filter based Healthy it is to be designed such that the prediction error must be
component Cancellation
minimum. The prediction error of the Wiener filter is defined
as the difference between actual components and predicted
Stage-3
Spectral Analysis of current signal components of stator current after de-noising and is given by
using Matrix Pencil Method

ed (n) = |id (n) − impd (n)| (3)

Calculation of Feature
Stage-4
Extraction parameter To reduce the prediction error in (3), minimum mean square
error technique is adapted to get filter coefficients of order p
Fault Detection Crite- [15]. From Fig.3, the mean square value of the error is given
Decision
ria based Fault Decision by

Fig. 1. Stage diagram of proposed fault detection criteria.

A. Stage-1(Wavelet De-Noising)
The wavelet de-noising of acquired stator current is done us-
ing Discrete Wavelet Transform (DWT) based decomposition.
The signal is decomposed into j number of levels and proper
threshold value is applied to remove noise. It is important that
the selection of number of levels to be decomposed is neither
too high nor too low. The j should be very precised such that it
only remove higher order noise but not fault components. Later Fig. 2. Block diagram of Wiener filter cancellation.
the resultant components are used to reconstruct the signal.

52
Authorized licensed use limited to: Rajamangala Univ of Technology Srivijaya provided by UniNet. Downloaded on August 31,2023 at 06:48:58 UTC from IEEE Xplore. Restrictions apply.
 C. Stage-3 (Matrix Pencil Method)
The equation (9) will contain the fault components along
with noise due to the process of pre-fault component can-
cellation. As the wavelet de-noising has reduced the severe
noise, the impact of noise existing after pre-fault component
cancellation is very less. Even this noise is low, the fault
component may be suppressed and will give wrong diagnosis
about fault. Therefore, the advanced spectral analysis tech-
nique namely matrix pencil method is employed to extract the
fault component. The method presented in [19] is adopted to
Fig. 3. Modified diagram of wiener filter cancellation. develop the algorithm for matrix pencil method. The signal
from equation (9) in time domain is expressed as

q
If a (n) = Imf sin(2πfm nTs ) (10)

p−1
m=1
ξ = E{|ed (n)|2 } = E{|id (n) − W (k)id (n − p0 − k)|2 } Where Ts is sampling period, q is the harmonic number of
k=0
(4) fault frequency, fm is the fault frequency and Im the magni-
Where E{|e(n)|2 } is expectation of mean square error and tude of fault component. Equation (10) may also expressed in
minimized by differentiate ξ with respect to W (k) and equat- terms of exponential component as
ing to zero and simplified to get filter coefficients as follows. 
M
If a (n) = Ri zin f or n = 0, 1, . . . N − 1 (11)
Ri W = rdi (5) i=1

Where zi = esi Ts = e(−αi +jωi )Ts

Where Ri is Toeplitz of auto correlation sequence, rdi cross Now the matrix pencil method is applied to equation (11)
correlation of desired signal and the observed signal and W is instead of (10) to reduce the complexity. As a process of
filter coefficients matrix [15]-[17]. solving MPM, the two matrices related to fault component
From the above equation, the filter coefficients can be can be expressed as
derived as
⎡ ⎤
if a1 if a2 ··· if aL
W (n) = Ri−1 (n)rdi (n) (6) ⎢ if a2 if a3 ··· if aL+1 ⎥
⎢ ⎥
[If a1 ] = ⎢ .. .. .. ⎥
Design of Wiener filter using minimum mean square sense ⎣ . . . ⎦
has shown poor performance especially at nascent stage of if aN −L if aN −L+1 ··· if aN −1 (N −L)×L
faults and can be improved using Kaiser window function (12)
which is given by and
 ⎡ ⎤
I0 [α 1−(
n− N
2
)2 ] if a0 if a1 ···
if aL−1
N
2 ⎢ if a1 if a2 if aL ⎥
···
K(n) = (7) ⎢ ⎥
I0 (β) [If a2 ] = ⎢ .. .. ⎥
..
⎣ . . ⎦
.
where I0 (x) is zeroth order Bessel function and β is the if aN −L−1 if aN −L · · · if aN −2 (N −L)×L
adjustable parameter. (13)
The kaiser window based Wiener filter coefficients can be The equations (12)-(13) can be rewritten as
obtained by multiplying (6) with (7) and is given below
[If a1 ] = [Z1 ][R][Z0 ][Z2 ] (14)
Wk (n) = W (n) ∗ K(n) (8)
[If a2 ] = [Z1 ][R][Z2 ] (15)
Pre-fault component cancellation of stator current using
modified filter coefficients can be obtained as shown in Fig.3. Where
The stator current after de-noising id (n) is convoluted with ⎡ ⎤
designed filter coefficients to get predicted components and 1 1 ··· 1
⎢ z1 z2 ··· zM ⎥
the resultant current with faulty components can be obtained ⎢ ⎥
[Z2 ] = ⎢ .. .. .. ⎥
as ⎣ . . . ⎦
z1N −L−1 z2N −L−1 ··· N −L−1
zM (N −L)×L
If a (n) = Id (n) − Impd (n) (9) (16)

53
Authorized licensed use limited to: Rajamangala Univ of Technology Srivijaya provided by UniNet. Downloaded on August 31,2023 at 06:48:58 UTC from IEEE Xplore. Restrictions apply.
 [Z0 ] = diag(z1 , z2 , · · · , zM ) (17)

[R] = diag(R1 , R2 , · · · , RM ) (18)

Where L and M are pencil parameter and order of the matrix
pencil respectively.
Then the magnitude and phase of each frequency compo-
nent existed in the stator current after pre- fault component
cancellation can be obtained by solving following equation.

⎡ ⎤ ⎡ ⎤⎡ ⎤
if a1 1 1 ··· 1 R1
⎢ if a2 ⎥ ⎢ z1 z2 ··· zM ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ R2 ⎥
⎢ .. ⎥=⎢ .. .. .. ⎥ ⎢ .. ⎥ (19) Fig. 4. Experimental arrangement.
⎣ . ⎦ ⎣ . . . ⎦⎣ . ⎦
if aN −L z1N −1 z2N −1 ··· N −1
zM RM
The experimental work is conducted by inserting faulty
Now the resultant signal after decomposing into vector bearing one by one into the three phase induction motor and
components using matrix pencil method is employed for stage the stator current is acquired for all cases. The process is
4 operation. executed for Outer race, Cage, Ball and generalized roughness
faults. The stator current under healthy and faulty conditions of
D. Stage-4 (Feature Extraction Parameter)
the bearing are acquired with 10.24kHz sampling frequency
Energy of the signal is to be calculated after spectral to get the required resolution in the spectrum and to match
analysis using matrix pencil method under healthy and faulty the number of samples and sampling frequency. The acquired
conditions using following expression. stator current is processed in LabVIEW using DAQ assistant
−1
and stored in offline. The stored stator current under both the

N
conditions of the bearing are used in MATLAB to execute the
E= |Imf a (n)|2 (20)
n=0
pre-written program. The obtained results with various stages
of proposed topoly is discussed in the following section.
Where Imf a (n) is stator current after 3 stage processing.
The fault indexing criteria is defined as the ratio of the energies IV. R ESULTS & D ISCUSSION
of faulty stator current to healthy stator current after 3 stage The stator current under various conditions of the bearing
processing and is given by are plotted in the Fig.5 and is observed that, the stator current
in time domain analysis doesn’t have any difference for healthy
Ef
σ= (21) and faulty condition which intern remains difficult to estimate
Eh the fault. In the time domain analysis stator current is acquired
Under normal operating condition the fault indexing param- under different conditions with sampling rate of 10.24 kHz and
eter σ shows the value of 1 which is the least value and a window of 4000 samples is shown in Fig.5. But the nature of
is increasing as fault grows up. The severity of the fault is the signal is remains same for all the conditions of the bearing.
estimated depending on the deviation of σ from 1. This leads to wrong diagnosis of the bearing health condition
which may cause sudden damage of the machine. Therefore,
III. E XPERIMENTAL VALIDATION the frequency domain analysis is carried out as mentioned in
The experimental validation of proposed topology of 4 stage section I.
processing of stator current to estimate the fault existence The periodogram of stator current under outer race fault
and its severity is shown in Fig.4. The experimental test bed condition for various stages is shown in Fig.6. The peri-
consists of a 3 phase induction motor with specifications: 2HP, odogram for direct acquired stator current, periodogram for
415 ± 10%V , 50 ± 5%Hz, Ambient Temperature of 400 C and pre-fault component canceled stator using Wiener filter and
degree of protection of IP55. The motor is exited through periodogram of stator current after matrix pencil method is
auto transformer by verifying with ammeter and voltmeter shown in Fig.6 respectively. In the fist stage, the domination
connected as shown in Fig.4. The motor is operated under of fundamental component, 7th and 11th harmonics along with
healthy and faulty conditions of the bearing and current is noise have suppressed the fault component along with few har-
extracted using current sensor of model LA 55P make by LEM monics too. In the second stage, the pre-fault components are
with conversion ratio of 1:1000. The primary and secondary canceled using Wiener filter. Due to the drawbacks associated
current ratings of 50A and 50mA respectively. The extracted with cancellation process, the fundamental component is still
stator current is processed into MATLAB by collecting sensed remain with less magnitude. In addition to this, the suppressed
current from current sensor through NI MyDAQ via LabVIEW. fifth harmonic is elevated as it is not visible in direct spectral

54
Authorized licensed use limited to: Rajamangala Univ of Technology Srivijaya provided by UniNet. Downloaded on August 31,2023 at 06:48:58 UTC from IEEE Xplore. Restrictions apply.
 Healthy Current

Magnitude (A)
4
2
0
-2
-4
0 500 1000 1500 2000 2500 3000 3500 4000
Time (Sec)
Magnitude (A) Magnitude (A) Magnitude (A) Magnitude (A)

Outerrace Fault
4
2
0
-2
-4
0 500 1000 1500 2000 2500 3000 3500 4000
Cage Fault Time (Sec)
4
2
0
-2
-4
0 500 1000 1500 2000 2500 3000 3500 4000
Time (Sec)
Ball Defect
4
2
0
-2
-4
0 500 1000 1500 2000 2500 3000 Time (Sec) 3500 4000
Generalized Roughness Fault
4
2
0
-2
-4
0 500 1000 1500 2000 2500 3000 3500 4000
Time (Sec)

Fig. 5. Stator current under various conditions of the bearing.

Power/frequency (dB/Hz) Power/frequency (dB/Hz) Power/frequency (dB/Hz)

Periodogram Power Spectral Density Estimate of direct current

50
Fundamental Component
0 Seventh Harmonic Eleventh Harmonic Noise Components

-50

-100
0 0.5 1 Frequency (kHz) 1.5 2 2.5
Periodogram Power Spectral Density Estimate of Wiener filter cancelled current
-100 Reduced Fundamental Component
Elevated Fifth Harmonic
-150 Removed Noise using Wavelet De-Noising

-200

-250
0 0.5 1 Frequency (kHz) 1.5 2 2.5

Periodogram Power Spectral Density Estimate of Matrix Pencil Method

120

100

80

60

40
0 0.5 1 1.5 2 2.5
Frequency (kHz)

Fig. 6. Periodogram of stator current under outer race fault with different stages of proposed topology.

55
Authorized licensed use limited to: Rajamangala Univ of Technology Srivijaya provided by UniNet. Downloaded on August 31,2023 at 06:48:58 UTC from IEEE Xplore. Restrictions apply.
 40 R EFERENCES
Direct MPencil
De-Noising based MPencil
35 Wiener Filetr & De-Noising based MPencil
[1] Liu, J., Wang, W., & Golnaraghi, F, “An extended wavelet spectrum for
30 bearing fault diagnostics,” IEEE Trans. Inst. Meas., vol. 57, no.12, pp.
25
2801-2812, July 2008.
Fault Severity

[2] Kompella, K. D., Mannam, V. G. R., & Rayapudi, S. R., “DWT based
20
bearing fault detection in induction motor using noise cancellation,”
15 Journal of Electrical Systems and Information Technology, vol.3, no.3,
10
pp. 411-427, 2016.
[3] Kompella, K. D., Rao, M. V. G., & Rao, R. S, “Bearing fault detection
5
in a 3 phase induction motor using stator current frequency spectral
0
Outerrace Cage Ball GR subtraction with various wavelet decomposition techniques, ” Ain Shams
Type of Fault
Engineering Journal, vol.9, no.4, pp. 2427-2439, 2018.
[4] A. Yazidi, H. Henao, G. A. Capolino, M. Artioli, and F. Filippetti,
Fig. 7. Bar graph for Fault indexing parameter “Improvement of frequency resolution for three-phase induction machine
fault diagnosis”, in Conf. Rec. IEEE IAS Annu, pp. 20-25, 2005.
[5] A. Bellini, A. Yazidi, F. Filippetti, C. Rossi, and G. Capolino, “High
frequency resolution techniques for rotor fault detection of induction
analysis. This may lead to inefficient fault detection and fault machines”, IEEE Trans. Ind. Electron., vol. 55, no. 12, pp. 4200-4209,
become dangerous. October 2008.
[6] A. G. Perez, R. J. R. Troncoso, E. C.Yepez, and R. A. O. Rios, “The
Therefore, in the 3rd stage, the pre-fault component can- application of high resolution spectral analysis for identifying multiple
celed stator current is used to decompose into eigenvalue combined faults in induction motors”, IEEE Trans. Ind. Electron., vol.
functions using proposed matrix pencil method mentioned 58, no. 5, pp. 2002-2010, April 2011.
[7] Climente-Alarcon, Vicente, et al.,“Induction motor diagnosis by ad-
in Section II. In this stage, all the frequency components vanced notch FIR filters and the Wigner Ville distribution”, IEEE Trans.
related to healthy current are removed as shown in Fig.6 and on Ind. Electron, vol. 61, no. 80, pp. 4217-4227, October 2013.
cause incorrect judgment that all the frequency components [8] Zeng-pu, Xu, Hu Li-xiang, and Qi Yu-ming, “Influence of window
width selection in fault diagnosis of loudspeaker based on Short-time
are removed. But the fault magnitude is mingled with noise Fourier Transform”,Proc. 3rd IEEE Int. Conf. Computer Science and
presented in spectrum of 3rd stage stator current. Moreover Information Technology , 2010.
estimation of severity of fault is also difficult due to hidden [9] W. Sun, G. Yang, Q. Chen, A. Palazoglu, and K. Feng, “Fault diagnosis
of rolling bearing based on wavelet transform and envelope spectrum
presence of fault frequency component. In this regard, the correlation”, J. Vib., Control, vol. 19, no. 6, pp. 924-941, April 2013.
stator current is processed for 4th stage, i.e feature extraction [10] K. V. Surti and C. A. Naik, “Bearing Condition Monitoring of Induction
parameter based fault estimation. In this stage, the fault Motor Based on Discrete Wavelet Transform & K-nearest Neighbor,”
2018 3rd International Conference for Convergence in Technology
indexing parameter, i.e energy of the stator current proposed (I2CT), Pune, 2018, pp. 1-5.
in Section II. The bar graph of fault indexing parameter for [11] Kompella, K. D., Rao, M. V. G., & Rao, R. S, “ SWT based bearing
various faults under three different categories are plotted in fault detection using frequency spectral subtraction of stator current with
and without an adaptive filter,” In TENCON 2017-2017 IEEE Region
Fig.7. At this stage, the fault component is clearly highlighted 10 Conference, pp. 2472-2477, November 2017.
by computing its fault indexing parameter after three stages [12] S. H. Kia, G. Cirrincione, H. Henao and G. -. Capolino, ”A computation-
of processing. This is clearly indicating that, fault has good ally efficient algorithm devoted to gear tooth localized fault detection in
induction machine-based systems,” 2016 XXII International Conference
indication only after 4 stages of processing and has unreliable on Electrical Machines (ICEM), Lausanne, 2016, pp. 2144-2150.
indication after direct and de-noising based Matrix pencil [13] E. H. E. Bouchikhi, V. Choqueuse and M. E. H. Benbouzid, “Current
method. From Fig.7. it is concluded that, the outer race fault Frequency Spectral Subtraction and Its Contribution to Induction Ma-
chines Bearings Condition Monitoring,” in IEEE Transactions on Energy
is very severe and may cause damage of the machine soon. Conversion, vol. 28, no. 1, pp. 135-144, March 2013.
[14] J. Zarei and S. Yousefizadeh, Fault detection in induction motors
using Parks vector approach and wavelet analysis, 2014 International
V. C ONCLUSION Conference on Mechatronics and Control (ICMC), Jinzhou, pp. 1064-
1068, 2014.
This paper addresses a novel approach to detect various [15] W. Zhou , B. Lu, T. G.Habetler, and R. G. Harley, “Incipient bearing
bearing faults experienced by three phase induction motor fault detection via motor stator current noise cancellation using wiener
using current signature analysis. The proposed topology has 4 filter,” IEEE Transactions on Industry Applications, vol. 44, no. 4, pp.
1309-1317, 2009.
stages of signal processing which include De-noising, Wiener [16] K. K. C. Deekshit , M. V. Rao, R. S. Rao, and R. Sreenivasu, “Estimation
filter based pre-fault component cancellation, Matrix Pen- of nascent stage bearing faults of induction motor by stator current
cil Method (MPM) based spectral analysis and computation signature using adaptive signal processing , in proc. IEEE INDICON,
pp. 1-5, 2013.
of feature extraction parameter as fault indication. Various [17] K. K. C. Deekshit, M. V. Rao, R. S. Rao, and R. Sreenivasu, “Estimation
bearing faults namely outer race, cage, ball and generalized of Bearing Faults in Induction motor by MCSA using daubechies
roughness have tested using proposed topology on 1.5 kW Wavelet Analysis ”, proc. IEEE ICSEG, pp.1-5, September 2014.
[18] S. Hedayati Kia, H. Henao, and Grard-Andr Capolino., “Diagnosis
three phase induction motor. The experimental results have of Broken-Bar Fault in Induction Machines Using Discrete Wavelet
pronounced the capability of proposed topology of fault de- Transform Without Slip Estimation,” IEEE Trans. Ind. Appl., vol. 45,
tection. Among the various faults tested, outer race fault has no. 4, pp. 1395-1404, July/August 2009.
[19] K. Sheshyekani, G. Fallahi, M. Hamzeh and M. Kheradmandi, ”A
severe impact and shows high fault indexing. In future, the General Noise-Resilient Technique Based on the Matrix Pencil Method
proposed topology may be applied for other faults like broken for the Assessment of Harmonics and Interharmonics in Power Systems,”
rotor, stator related and eccentricity faults and may test under in IEEE Trans. on Power Delivery, vol. 32, no. 5, pp. 2179-2188, October
2017.
variable load condition.

56
Authorized licensed use limited to: Rajamangala Univ of Technology Srivijaya provided by UniNet. Downloaded on August 31,2023 at 06:48:58 UTC from IEEE Xplore. Restrictions apply.