Article

Journal of Vibration and Control

18(11) 1585–1594

Application of high-resolution spectral ! The Author(s) 2011

Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
analysis for identifying faults in induction DOI: 10.1177/1077546311422925
jvc.sagepub.com
motors by means of sound

Arturo Garcia-Perez1, Rene J Romero-Troncoso1,

Eduardo Cabal-Yepez1, Roque A Osornio-Rios2
and Jose A Lucio-Martinez1

Abstract
Induction motors are critical components for most industries. Induction motor failures may yield an unexpected inter-
ruption at the industry plant. Several conventional vibration and current analysis techniques exist by which certain faults
in rotating machinery can be identified. Ever since the first motor was built, plant personnel have listened to the noises
emanating from machines; with enough experience, a listener may make a fairly accurate estimate of the condition of a
machine. Although there are several works that deal with vibration and current analysis for monitoring and detection of
faults in induction motors, the analysis of sound signals has not been sufficiently explored as an alternative non-invasive
monitoring technique. The contribution of this investigation is the development of a condition monitoring strategy than
can make a reliable assessment of the presence of specific fault condition in an induction motor with a single fault present
through the analysis of a sound signal. The proposed methodology is based on the multiple-signal classification algorithm
for high-resolution spectral analysis. Results show that the proposed methodology of sound analysis could improve
standard techniques for induction motor fault detection, enhancing detectability.

Keywords
Fault detection, induction motor, sound, spectral analysis, vibration
Received: 15 October 2010; accepted: 27 July 2011

1. Introduction stator, the rotor, and the bearings (Benbouzid, 2000).

Induction motors are widely used and they are consid- Traditionally, the condition monitoring of induction
ered critical components for electric utilities and pro- machines has been divided into two areas: mechanical
cess industries (Siyambalapitiya and McLaren, 1990). problems and electrical problems (Al Kazzaz and
An induction motor failure may yield an unexpected Singh, 2003). Mechanical problems include bearing
interruption at the industry plant, with consequences wear, rotor unbalance, and airgap distortion.
in costs, product quality, and safety. These faults may
be inherent in the machine itself or in operating condi-
tions. The origins of inherent faults are due to the 1
HSPdigital – CA Telematica, DICIS, Universidad de Guanajuato,
mechanical or electrical forces acting on the machine. Salamanca, Gto., Mexico
2
The greatest challenge in the area of condition moni- HSPdigital – CA Mecatronica, Facultad de Ingenieria, Universidad
Autonoma de Queretaro, San Juan del Rio, Qro., Mexico
toring is the diagnosis of a fault before it becomes crit-
ical and an early detection of this allows for the repair Corresponding author:
RJ Romero-Troncoso, HSPdigital – CA Telematica, DICIS, Universidad de
of the fault. In general, condition-monitoring schemes Guanajuato, Carr. Salamanca-Valle km 3.5þ1.8, Palo Blanco, 36700
have been widely used to sense specific failure modes in Salamanca, Gto., Mexico
one of three main induction motor components: the Email: troncoso@hspdigital.org

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

1586 Journal of Vibration and Control 18(11)

Electrical problems include broken rotor bars and Sadeghian et al. (2009) present an algorithm for
winding insulation breakdown. Typically, accelerome- online detection of rotor bar breakage through the
ters are used to measure mechanical vibrations for the use of wavelets and neural networks. Chen and
detection of mechanical faults, and current probes are Zivanovic (2010) make the estimation of frequency
used to monitor electrically related problems before components in a stator current for the detection of
catastrophic failures in a component occur. Almost broken rotor bars in induction machines. Zarei and
40% to 50% of all failures are bearing related, Poshtan (2009) present a bearing fault detection using
around 5% to 10% are rotor faults, and unbalance a wavelet packet transform of the induction motor
faults are within the 12% of other faults. Vibration stator current. Teotrakool at al. (2009) present the
and current analysis have long been used for the detec- application of motor current signature analysis using
tion and identification of machine fault conditions. The wavelet packet decomposition to detect bearing faults.
specific characteristics of the vibration and current Eren and Devaney (2004) present a method for detect-
spectra that are associated with common fault condi- ing a motor bearing fault using wavelets. Kau and
tions are well known. For instance, the ball pass outer Ngan (2010) present the detection of a motor bearing
raceway frequency (fBPOF) is the spectral component of outer raceway defect using wavelets. Pineda-Sanchez
the ball pass outer raceway, reflecting bearing defects; et al. (2010) presents a diagnosis of broken rotor bars
the peak at the rotational frequency in the vibration using the fractional Fourier domain. However, other
spectrum, indicating the degree of unbalance; and the techniques are focused in current and vibration analy-
slip frequency sidebands around the fundamental fre- sis; for example, Rangel-Magdaleno et al. (2009) pre-
quency, indicating broken rotor bars.The presence of sent a methodology for on-line half broken rotor bars,
these deteriorating fault conditions determines when a which combines current and vibrations analysis by cor-
machine should be taken out of operation. relating the signal spectra. Kar and Mohanty (2008)
Many different techniques have been proposed for present an experimental investigation of fault diagnosis
the surveillance and diagnosis of the rotating machin- in a multistage gearbox under transient loads using
ery in the literature, some of them are focused in vibra- vibration and current signals. Bellini et al. (2006) pre-
tion analysis; for example, Contreras-Medina et al. sent a multi-variable monitoring system for on-line
(2010) present an FPGA-based multiple channel vibra- monitoring of induction motors using vibrations, cur-
tion analyzer for detection of broken rotor bars, unbal- rent and stray flux signals. Hao and Chu (2009) present
ance and looseness. Garcia-Perez et al. (2011) make the a scheme with morphological undecimated wavelet
detection of broken rotor bars, unbalance and bearing decomposition, where the noise produced from a roll-
faults using high frequency resolution using the multi- ing element bearing defect is analyzed for fault diagnos-
ple-signal classification (MUSIC) algorithm. Rangel- tics. Wang and Chang (2005) present an application of
Magdaleno et al. (2010) present a vibration analyzer the MUSIC algorithm applied to the identification of
for continuous CNC machinery monitoring with the number and localization of acoustic sources.
fused fast Fourier transform-discrete wavelet transform Sarradj (2010) presents an application of the MUSIC
(FFT-DWT). Patel and Darpe (2009) present a vibra- algorithm applied to phased microphone array for cor-
tion signature analysis of rotor with rotor-stator rub, rect quantitative estimation of acoustic source spectra.
transverse fatigue crack and unbalance. Babu and Nagata et al. (2009) present an application of MUSIC
Sekhar (2008) make the detection of two cracks in a applied to the problem of direction-of-arrival estima-
rotor-bearing system using amplitude deviation curve tion both azimuthal and elevation form binaural sound.
technique. Jun and Gadala (2008) present the analysis Ever since the first motor was built, plant personnel
of dynamic behavior of a cracked rotor. Yan and Gao have listened to the noises emanating from machines;
(2009) present a filter construction technique for with enough experience, a listener may make a fairly
enhanced defect identification in rotary machine sys- accurate estimate of the condition of a machine. At
tems. Jalan and Mohanty (2009) present a model best, this is merely a qualitative analysis; at worst it
based technique for fault diagnosis of rotor-bearing may give an entirely wrong assessment. Meanwhile,
systems for misalignment and unbalance. Liu et al. very few works deal with sound signals for the diagno-
(2010) present a diagnostic scheme for bearing fault sis and identification of faults on induction motor sys-
diagnostics using neurofuzzy classifier. Yan and Gao tems. Wu and Liao (2010) present a neural network
(2010) present an energy based approach to detect bear- system for automotive air-conditioner blower fault
ing defects in rotary machines. Other techniques are diagnosis using noise emission signal. Tinta et al.
focused in current analysis; for example, Ordaz- (2005) present a technique that provide reliable esti-
Moreno et al. (2008) present the detection of broken mate of fault diagnosis of vacuum cleaner motors.
rotor bars on induction motors based on the DWT Wang et al. (2005) present a numerical simulation for
application during the start-up current transient. predicting the sound power from an inverter-driven

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

Garcia-Perez et al. 1587

induction motor. The analysis of machine noise begins barely visible. If a motor is out of balance or misa-
with an understanding of the possible causes of the ligned, the signature of unbalance is a vibration
noise. In an induction motor, these include: slot har- signal, which normally has the form of an increased
monics, supply harmonics, rotor unbalance, winding amplitude along the rotating frequency and its har-
asymmetry and the bearings. Some of these are inherent monics (Bloch, 1999). Vibration analysis can provide
in the design of the machine and hence should remain a quick and relatively easy way to extract information
constant over its lifetime (slot harmonics, winding that may relate the unbalance fault in an induction
asymmetry). Some develop over time, and may lead motor.
ultimately to machine failure (bearing, unbalance,
broken rotor bars). Others are due to external circum-
stances (supply harmonics). All of these add to the
2.2. Bearings
sound emitted by a machine, resulting in a very com- McFadden and Smith (1985) give a review of the causes
plex audio signal. This work investigates a method that and expected frequencies of vibration due to rolling
may be used to measure and analyze the sound of an element bearings. From the geometry of the bearing,
induction machine, where the feasibility of distinguish- various theoretical frequencies can be calculated such
ing between broken rotor bars, bearing and unbalance as the inner and outer race element pass frequencies,
defects are the objectives of this investigation. cage rotational frequency and rolling element spin fre-
The contribution of this investigation is the study quency. A defect on the outer race will cause an impulse
for developing a condition monitoring strategy than each time a rolling element contacts the defect. The
can make a reliable assessment of the presence of a rotor speed (fr) is the frequency at which the inner race-
specific fault condition in an induction motor with a way rotates, that must be the frequency of the shaft.
single fault present through the analysis of a sound The theoretical vibration frequency of the ball pass
signal. Combined with vibration analysis, it is outer raceway frequency (fBPOF) is easily determined as:
believed that a more accurate diagnostic system  
can be designed. Yet, the sound signal analysis by NB Db
fBPOF ¼ fr 1  cos  ð1Þ
itself has shown excellent results in the identification 2 Dc
of bearing, unbalance and broken rotor bar faults in
an induction motor. where  is the contact angle between the bearing sur-
The proposed method uses a high-resolution spectral face, Dc is the cage diameter of the bearing and is mea-
analysis based on the MUSIC algorithm, which is sured from a ball center to the opposite ball center, Db
applied to the sound signal produced by an induction is the ball diameter and NB is the number of balls in the
motor for identification of the frequency related fault. bearing.
In this research, three different faults in an induction
motor such as broken rotor bars, unbalance, and bear-
ings are investigated in an experimental way. Also, the
2.3. Broken rotor bars
MUSIC algorithm is applied to the vibrations signals
for validating results, obtained from the sound analysis. An induction motor operating with a broken rotor bar
Results show the potentiality of the methodology as a defect creates a negative sequence of rotor current due
deterministic detection technique that is suited for the to rotor asymmetry. It induces a principal component
detection of broken rotor bars, bearing and unbalance in the spectrum of stator current, which rises up to the
faults in induction motors. frequency (1–2s)fs and it is close to the fundamental
frequency. Due to reflection, the rotor asymmetry fre-
quencies are 2ksfs , where k ¼ 1,2,3,. . . is any positive
integer and the s is the slip. The slip frequency (fslip) is
2. Fault-related components
the difference between synchronous speed and rotor
Three different induction motor faults are considered in speed. Thus, in the case of broken rotor bars, there is
the paper: unbalance, bearing and broken rotor bars. a speed oscillation, which acts as a frequency modula-
tion on rotation frequency and two side band frequen-
cies appear around fr in vibration spectrum (Sadoughi
2.1. Unbalance et al., 2006) and is given by
The number of poles determines the speed of an induc-
tion motor and the speed of the motor can be identified fBRB ¼ fr  2ksfS ð2Þ
by a peak in the spectrum and then monitored at
changes in amplitude. A properly balanced and aligned where fBRB are the sideband frequencies associated with
motor has a frequency peak related to its speed that is the broken rotor bar.

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

1588 Journal of Vibration and Control 18(11)

is the vector of complex sinusoids, and B defines the

3. Theoretical background power of the sound or vibration space vector as:
3.1. MUSIC algorithm applied to sound  
B ¼ aaH ¼ diag I21 I22    I2P ð7Þ
or vibration space vector
The MUSIC algorithm estimates the frequencies of the Now, consider an eigen-decomposition of the corre-
complex sinusoids that best approximate a noisy signal lation matrix RiS that has only P nonzero eigenvalues,
by using an eigen-based decomposition method and is given by
(Garcia-Perez et al., 2011; Kia et al., 2007). First, con-
sider the sound or vibration signal i(t) as a sum of X
N X
P

P complex sinusoids and white noise: RiS ¼ j gj gH

j ¼ j gj gH
j ð8Þ
j¼1 j¼1

X
P
iðtÞ ¼ Ik ej ð2fk tþk Þ þ in ðtÞ ð3Þ where j and gj are the eigenvalues and eigenvectors of
k¼1 the matrix RiS , respectively. The sum of the cross prod-
ucts of the eigenvectors forms the identity matrix, and it
is possible to rewrite the autocorrelation matrix of the
where Ik , fk , and k are the amplitude, the frequency noise Rin as stated in (9).
and the phase of the k-th sound or vibration space
vector, respectively, and in ðtÞ is white noise. An X
N

N-sample vector i ¼ ½iðmÞ, iðm þ 1Þ, . . . , iðm þ N  1Þ Rin ¼ n2 I ¼ n2 gk gH
k ð9Þ
k¼1
of the noisy signal, sampled at frequency fs, can be
written as:
From (8) and (9), it is possible to rewrite the auto-
correlation matrix of the noisy signal as:
i ¼ iS þ in ¼ Sa þ in ð4Þ
X
P X
N
2
where the signal vector iS ¼ Sa is defined as: Ri ¼  k gk gH
k þ n gk gH
k
k¼1 k¼1
2 3 ð10Þ
ej2F1 m e j2F2 m  e j2FP m X
P   X
N

6 e j2F1 ðmþ1Þ 7 ¼ k þ n2 gk gH

k þ n
2
gk gH
k
6 e j2F2 ðmþ1Þ  ej2FP ðmþ1Þ 7 k¼1 k¼1þP
iS ¼ 6
6 .. .. .. .. 7
7
4 . . . . 5 The eigenvectors and eigenvalues of the correlation
e j2F1 ðmþN1Þ e j2F2 ðmþN1Þ    ej2FP ðmþN1Þ matrix Ri can be divided in two disjoint groups. The
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
S group of eigenvectors associated with the largest P
2 3 eigenvalues span of the signal subspace and the group
I1 ej1
6 I ej2 7 of noise with an eigenvalue given by n2 . Therefore, the
6 2 7 signals vectors can be expressed as a linear combination
6 7
6 .. 7 of the P-eigenvectors span of the signal. The MUSIC
4 . 5
pseudo-spectrum Q of the sound or vibration space
IP ejP
|fflfflfflfflfflffl{zfflfflfflfflfflffl} vector is given by:
a
ð5Þ 1
QMUSIC
i ðFÞ ¼ ð11Þ
P
N
sH ðFÞgk 2
where Fk ¼ fk =fs . The autocorrelation matrix Ri of the k¼Pþ1
noisy signal i is the sum of the autocorrelation matrices
of the signal iS and the noise in (RiS and Rin respectively) Expression (11) exhibits the peaks that are exactly at
is given by the frequencies of the principal sinusoidal components
where sHk ðFk Þgk ¼ 0:
X
P
Ri ¼ RiS þ Rin ¼ SBSH þ n2 I ¼ I2k sk sH 2
k þ n I ð6Þ
k¼1 3.2. Sound analysis
Sound signals recorded in a noisy industrial environ-
where the exponent H denotes
 Hermitian transpose, I is ment by using partial sound protection can still be uti-
the identity matrix, sH
k ¼ 1 e j2Fk
   ej2Fk ðN1Þ lized for feature extraction. Namely, the anomalies in

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

Garcia-Perez et al. 1589

the spectrum of the sound signals caused by the motor

fault do not overlap with those caused by environmen-
tal noise (Benkoa et al., 2003). Li and Mechefske (2006)
present the Wigner-Ville distribution technique applied
for identification of a broken rotor bar and a combina-
tion of bearing faults using current, vibration and
acoustic methods. Various faults can be recognized
from the sound signal in harmonics related to the
faults’ frequencies, e.g. damages in bearing, unbalance
and broken rotor bars. Tinta et al. (2005) have shown
the study of a faulty motor, where the power-spectral-
density (PSD) at certain characteristic frequencies
strongly increase. Vibrations of solid structures and
mechanical contact between different surfaces of the
motor components cause mechanical noise. Vibrations
arise from unbalanced rotating parts, where these
vibrations increase at higher rotational speeds or
when a fault occurs in the induction motor. The vibra-
tional waves propagate all over the housing, thus cre-
ating reverberant vibrational fields that radiate sound.
A fault-free motor is significantly less noisy than a
faulty motor, which increases substantially the intensity
of mechanical noise in the lower range of rotational
speeds and reaches much higher values. It is expected
to find frequencies and harmonics in the spectral anal-
ysis of sound signals related to the different vibration
faults in an induction motor.

4. Experimental setup Figure 1. (a) Test bench used during the experiment. (b) Block
diagram of the proposed multiple fault diagnosis system for
A sound signal is captured in order to identify the
induction motors.
dynamic characteristics of the induction motor; and,
simultaneously, vibration signals are also acquired in
order to verify the obtained results. Forty tests were the signal corresponding to the x axis (AX). The
made for sound and vibration on each condition in audio signal is amplified utilizing a power amp
order to validate the proposed methodology. Figure 1 Marshall model MG15CDR. A 12-bit 4-channel
shows the experiment setup where different three-phase serial-output analog to digital converter (ADC)
induction motors (model WEG 00136APE48T, 740 (ADS7841) is used for data acquisition of audio and
Watts) are used to test the performance of the proposed vibrations. The instrumentation system uses a sampling
methodology identifying the fault conditions treated in frequency of 4 kHz for obtaining 32768 samples of
this work. The tested motors have two poles, 28 bars audio whereas 4096 samples are obtained at a sampling
and receive a power supply of 220 VAC at 60 Hz, and rate of 1500 Hz from vibrations signals during the
the applied mechanical load is that of an ordinary alter- induction motor steady state. Figure 1(b) shows the
nator. The audio signal is acquired using a condenser proposed multiple fault diagnosis system for induction
microphone JST model CX-509, which has a cardioid motor monitoring utilizing audio signals.
polarization pattern. The microphone was placed in a Three different fault conditions are treated in this
convenient position where the motor vibration did not work:
have any influence in the microphone. However, for
different microphone positions; the obtained results
show some degree of variation. On the other hand,
4.1. Broken rotor fault
the vibration signal is acquired using a MEMS-based The broken bar condition is produced artificially by
triaxial accelerometer (model LIS3L02AS4) from drilling two holes with 7.938 mm of diameter without
STMicroelectronics placed as shown in Figure 1(a). harming the rotor shaft. Figure 2(a) shows the utilized
From the three acceleration axes (x, y and z) of the rotor with two broken rotor bars. The motor is running
vibration signal, the best results were obtained with at 3,465 RPM in a 60 Hz system; the tested induction

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

1590 Journal of Vibration and Control 18(11)

Figure 2. Artificially generated faults (a) Two broken bar, (b) Outer race damaged bearing, (c) Unbalanced pulley.

motor has a rotor frequency fr ¼ 57.75 Hz. Thus, the bearing defect and unbalance) is present in the induc-
slip s is 0.0356 and the sideband frequencies associated tion motor. In order to verify the validity of the sound
with the broken rotor bar are calculated by using equa- analysis, it is compared against the analysis of the
tion (2) (Sadoughi et al., 2006), resulting in 53.5 Hz vibration signals for the studied single faults. The anal-
and 62.1 Hz. ysis of sound and vibration signals is implemented in
the Matlab Digital Signal Processing Toolbox. The
MUSIC algorithm has an order of 14.
4.2. Bearing fault
To carry out the faulty bearing test, the bearing is arti-
5.1. Analysis of broken rotor bar fault
ficially damaged by drilling two holes with 1.191 mm of
diameter on its outer race using a tungsten drill bit. The broken rotor bar fault is first, identified in vibra-
Figure 2(b) shows the artificially damaged bearing tion signal with the presence of 53.5 Hz and
6203-2ZNR used in this work. The vibration character- 62.1 Hz frequency components in the spectrum as
istic defect frequency of the rolling element bearing described in section 4. Figure 3 shows the power-spec-
outer race is calculated by using equation (1) (Nagata tral-density obtained with the MUSIC algorithm,
et al., 2009). The tested induction motor has a rotor where the observed frequencies are very close to the
frequency fr ¼ 57.75 Hz and a test bearing having eight analytical frequencies for the case of two broken
balls of diameter 6.5 mm and the cage diameter of the rotor bars fault.
bearing of 28 mm with contact angle b ¼ 0, thus, the The broken rotor bar fault is also clearly identified in
ball pass outer raceway frequency defect (fBPOF) is the sound signal with the presence of twelfth-order har-
found to be 177.7 Hz. monics of the side band frequencies (fBRB), that are
located at 688 Hz and 694 Hz in the spectrum obtained
with MUSIC algorithm, as shown in Figure 4.
4.3. Unbalance fault
The unbalance condition is present when the induction
5.2. Analysis of bearing fault
motor mechanical load is not uniformly distributed,
taking the center of mass out of the motor shaft. The detection of a bearing defect is first identified in
Figure 2(c) shows a pulley with an added mass used vibrations signal where the ball pass outer raceway fre-
for generating unbalance on the induction motor. The quency defect is found (fBPOF ¼ 177.7 Hz) as described
signature of unbalance in a vibration signal normally in section 4. Figure 5 shows the power-spectral-density
has the form of an increased amplitude along the rotor obtained with the MUSIC algorithm, where the identi-
frequency, being located in this case at 57.75 Hz. fication of the bearing fault is clearly detected.
The bearing fault is also clearly identified in the
sound signal with the presence of the third-order har-
5. Experiment results
monic of the ball pass outer raceway frequency (fBPOF)
The proposed sound analysis has been applied to sev- at 533 Hz in the spectrum obtained with the MUSIC
eral cases in which a single fault (broken rotor bar, algorithm, as shown in Figure 6.

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

Garcia-Perez et al. 1591

Figure 3. Analysis regions of vibration signal for (a) a healthy Figure 5. Analysis regions of vibration signal for (a) a healthy
motor and (b) a motor with two broken rotor bars. motor and (b) a motor with bearing defect.

Figure 6. Analysis regions of sound signal for (a) a healthy

Figure 4. Analysis regions of sound signal for (a) a healthy motor and (b) a motor with bearing fault.
motor and (b) a motor with two broken rotor bars.

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

1592 Journal of Vibration and Control 18(11)

Table 1. Detectability in decibels for the faults analysis

Condition FFT Wigner-Ville MUSIC

Broken rotor bar 1 5 20

Bearing defect 0 15 20
Unbalance 8 30
FFT: Fast Fourier Transform, MUSIC: Multiple-signal classification
algorithm.

5.3. Analysis of unbalance fault

The unbalance defect is verified with the increasing
value of the rotor frequency fr ¼ 57.75 Hz in the vibra-
tion spectrum as described in section 4. Figure 7 shows
the power-spectral-density obtained with the MUSIC
algorithm, where the identification of the unbalance
fault is clearly detected. This fault is clearly identified
in the sound signal spectrum with the presence of sev-
eral harmonic component frequencies (sixth, seventh,
eighth, ninth and tenth-order) located at 346.5,
404.25, 462, 520 and 577 Hz in the spectrum obtained
with the MUSIC algorithm as shown in Figure 8.
Table 1 shows the detectability comparison in dB for
the proposed methodology against the traditional FFT
Figure 7. Analysis regions of vibration signal for (a) a healthy and the Wigner-Ville methods (Li and Mechefske,
motor and (b) a motor with unbalance defect. 2006) where it is noticeable that for most cases the
detectability is improved over 20 dB in the proposed
methodology. The detectability is calculated as the
amplitude ratio (in dB) between the faulty over the
healthy condition.

6. Conclusion
This paper proposes the identification of induction
motor faults by the application of the MUSIC algo-
rithm to sound signals. It has been shown that the
power-spectral-density of sound signals for broken
rotor bars, bearing, and unbalance faults exhibit har-
monic frequencies, related to the different vibrations
faults in an induction motor. The validation of the pro-
posed technique is done through the comparison
against the analysis of vibration signal, an already
accepted methodology for induction motor fault mon-
itoring and detection. Thus, standard techniques for
induction motor fault detection can be improved with
the proposed methodology of sound analysis, enhanc-
ing detectability and also benefit from the non-invasive
nature of the sound monitoring. However, for greater
development, there is still the necessity of making fur-
ther investigations to find the best microphone position,
or to use several microphones placed in different posi-
Figure 8. Analysis regions of sound signal for (a) a healthy tions and with special algorithms for noise cancellation
motor and (b) a motor with unbalance fault. to consider as an option for future investigations.

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

Garcia-Perez et al. 1593

Funding multiresolution Fourier transform. Journal of Sound and

This research received no specific grant from any funding Vibration 311(1): 109–132.
agency in the public, commercial, or not-for-profit sectors. Kau ECCL and Ngan HW (2010) Detection of motor bearing
outer raceway defect by wavelet packet transformed motor
current signature analysis. IEEE Transactions on
References Instrumentation Measurement 59(10): 2683–2690.
Al Kazzaz SAS and Singh GK (2003) Experimental investi- Kia SH, Henao H and Capolino GA (2007) A high-resolution
gations on induction machine condition monitoring and frequency estimation method for three-phase induction
fault diagnosis using digital signal processing techniques. machine fault detection. IEEE Transactions on Industrial
Electric Power Systems Research 65(3): 197–221. Electronics 54(4): 2305–2314.
Babu TR and Sekhar AS (2008) Detection of two cracks in a Li E and Mechefske C (2006) Detection of induction motor
rotor-bearing system using amplitude deviation curve. faults: a comparison of stator current, vibration and
Journal of Sound and Vibration 314(3): 457–464. acoustic methods. Journal of Vibration and Control 12(2):
Bellini A, Concari C, Franceschini G, Tassoni C and Toscani 165–188.
A (2006) Vibrations, currents and stray flux signals to Liu J, Wang W and Golnaraghi F (2010) An enhanced diag-
assess induction motors rotor conditions. In Proceedings nostic scheme for bearing condition monitoring. IEEE
of 32nd Annual Conference on IEEE Industrial Electronics, Transactions on Instrumentation Measurement 59(2):
Paris, France, Nov. 6–10, pp. 4963–4968. 309–321.
Benbouzid MEH (2000) A review of induction motors signa- McFadden PD and Smith JD (1985) The vibration produced
ture analysis as a medium for faults detection. IEEE by multiple points defects in a rolling element bearing.
Transactions on Industrial Electronics 47(5): 984–993. Journal of Sound and Vibration 98(2): 263–273.
Benkoa U, Petrovcic J, Juricic D, Tavcar J, Rejec J and Nagata Y, Iwasaki S, Hariyama T, Fujioka T, Obara T and
Stefanovska A (2003) Fault diagnosis of a vacuum cleaner
Wakatake T (2009) Binaural localization based on
motor by means of sound analysis. Journal of Sound and
weighted wiener gain improved by incremental source
Vibration 276(3): 781–806.
attenuation. IEEE Transactions on Audio, Speech, and
Bloch HP (1999) Vibration Analysis in Machinery Failure
Language Processing 17(1): 52–65.
Analysis and Troubleshooting. Houston, TX: Gulf
Ordaz-Moreno A, Romero-Troncoso RJ, Vite-Frias JA,
Professional Publishing.
Rivera-Gillen JR and Garcia-Perez A (2008) Automatic
Chen S and Zivanovic R (2010) Estimation of frequency com-
online diagnosis algorithm for broken-bar detection on
ponents in stator current for the detection of broken rotor
induction motors based on discrete wavelet transform
bars in induction machines. Journal of Measurement 43(7):
for FPGA implementation. IEEE Transactions on
887–900.
Industrial Electronics 55(5): 2193–2202.
Contreras-Medina LM, Romero-Troncoso RJ, Cabal-Yepez
Patel TH and Darpe AK (2009) Coupled bending-torsional
E, Rangel-Magdaleno JJ and Millan-Almaraz JR (2010)
FPGA-based multiple-channel vibration analyzer for vibration analysis of rotor with rub and crack. Journal of
industrial applications in induction motor failure detec- Sound and Vibration 326(5): 740–752.
tion. IEEE Transactions on Instrumentation Measurement Pinenda-Sanchez M, Riera-Guasp M, Antonino-Daviu JA,
59(1): 63–72. Roger-Folch J, Perez-Cruz J and Puche-Panadero R
Eren L and Devaney MJ (2004) Bearing damage detection via (2010) Diagnosis of induction motor faults in the frac-
wavelet packet decomposition of the stator current. IEEE tional Fourier domain. IEEE Transactions on
Transactions on Instrumentation Measurement 53(2): Instrumentation Measurement 59(8): 2065–2075.
431–436. Rangel-Magdaleno JJ, Romero-Troncoso RJ, Osornio-Rios
Garcia-Perez A, Romero-Troncoso RJ, Cabal-Yepez E and RA, Cabal-Yepez E and Contreras-Medina LM (2009)
Osornio-Rios RA (2011) The application of high-resolu- Novel methodology for online half-broken-bar detection
tion spectral analysis for identifying multiple combined on induction motors. IEEE Transactions on
faults in induction motors. IEEE Transactions on Instrumentation Measurement 58(5): 1690–1698.
Industrial Electronics 58(5): 2002–2210. Rangel-Magdaleno JJ, Romero-Troncoso RJ, Osornio-Rios
Hao R and Chu F (2009) Morphological undecimated wave- RA, Cabal-Yepez E and Dominguez-Gonzalez A (2010)
let decomposition for fault diagnostics of rolling element FPGA-based vibration analyzer for continuous CNC
bearings. Journal of Sound and Vibration 320(4): machinery monitoring with fused FFT-DWT signal pro-
1164–1177. cessing. IEEE Transactions on Instrumentation
Jalan AK and Mohanty AR (2009) Model based fault diag- Measurement 59(12): 3184–3194.
nosis of a rotor–bearing system for misalignment and Sadeghian A, Ye Z and Wu B (2009) Online detection of
unbalance under steady-state condition. Journal of Sound broken rotor bars in induction motors by wavelet packet
and Vibration 327(3): 604–622. decomposition and artificial neural networks. IEEE
Jun OS and Gadala MS (2008) Dynamic behavior analysis of Transactions on Instrumentation Measurement 58(7):
cracked rotor. Journal of Sound and Vibration 309(1): 2253–2263.
210–245. Sadoughi A, Ebrahimi M and Rezaei E (2006) A new
Kar C and Mohanty AR (2008) Vibration and current tran- approach for induction motor broken bar diagnosis by
sient monitoring for gearbox fault detection using using vibration spectrum. In Proceedings of International

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015

1594 Journal of Vibration and Control 18(11)

Joint Conference SICE-ICASE, Busan, Korea, Oct. 18–21, experimental investigation. IEEE Electric Power
pp. 4715–4720. Applications 149(1): 46–52.
Sarradj E (2010) A fast signal subspace approach for the Wang JH and Chang C (2005) Identification of the number
determination of absolute levels from phased microphone and locations of acoustic sources. Journal of Sound and
array measurements. Journal of Sound and Vibration Vibration 284(7): 393–420.
329(9): 1553–1569. Wu J and Liao S (2010) Fault diagnosis of an automotive air-
Siyambalapitiya DJ and McLaren PG (1990) Reliability conditioner blower using noise emission signal. Expert
improvement and economic benefits of on-line monitoring Systems with Applications 37(2): 1438–1445.
systems for large induction machines. IEEE Transactions Yan R and Gao R (2009) Energy-based feature extraction for
on Industrial Applications 26(6): 1018–1025. defect diagnosis in rotary machines. IEEE Transactions on
Teotrakool K, Devaney MJ and Eren L (2009) Adjustable- Instrumentation Measurement 58(9): 3130–3139.
speed drive bearing-fault detection via wavelet packet Yan R and Gao R (2010) Harmonic wavelet-based data fil-
decomposition. IEEE Transactions on Instrumentation tering for enhanced machine defect identification. Journal
Measurement 58(8): 2747–2754. of Sound and Vibration 329(15): 3203–3217.
Tinta D, Petrovcic J, Benko U, Juricic D, Rakar A and Zele Zarei J and Poshtan J (2009) An advanced Park’s vectors
M (2005) Fault diagnosis of vacuum cleaner motors. approach for bearing fault detection. Tribology
Control Engineering Practice 13(2): 177–187. International 42(2): 213–219.
Wang C, Astfalck A and Lai JCS (2005) Sound power radi-
ated from an inverter-driven induction motor:

Downloaded from jvc.sagepub.com at CAMBRIDGE UNIV LIBRARY on January 11, 2015