Spectral Analysis for Identifying Faults in Induction

Motors by Means of Sound

Fernando Salazar-Villanueva, Oscar G. Ibarra-Manzano

Universidad de Guanajuato.
Comunidad de Palo Blanco s/n, C.P. 36885
Salamanca, Guanajuato, Mexico
f.salazarvillanueva@ugto.mx, ibarrao@ugto.mx

Abstract—Induction motors are critical components for most insulation breakdown. Typically, accelerometers are used to
industries. Induction motors failures may yield an unexpected measure mechanical vibrations for the detection of mechanical
interruption at the industry plant. Several conventional vibration faults, and current probes are used to monitor electrically
and current analysis techniques exist by which certain faults in related problems before catastrophic failures in a component
rotating machinery can be identified. Ever since the first motor occur. Almost 40%-50% of all failures are bearing related,
was built, plant personnel have listened to the noises emanating around 5%-10% are rotor faults, and unbalance faults are
from machines; with enough experience, a listener may make a within the 12% of others faults. Vibration and current analysis
fairly accurate estimate of the condition of a machine. Although have long been used for the detection and identification of
there are several works that deal with vibration and current
machine fault conditions. The specific characteristics of the
analysis for monitoring and detection of faults in induction
vibration and current spectra that are associated with common
motors, the analysis of sound signals has not been sufficiently
explored as an alternative non-invasive monitoring technique. fault conditions are well known.
The contribution of this investigation is the development of a Many different techniques have been proposed for the
condition monitoring strategy that can make reliable assessment surveillance and diagnosis of the rotating machinery in
of the presence of a specific fault condition in an induction motor literature, some of them are focused in vibration analysis; for
with a single fault present through the analysis of sound signal. example, Babu and Sekhar [3] make the detection of two
The proposed methodology is based on the combination of cracks in a rotor-bearing system using amplitude deviation
Intrinsic Mode Functions (IMFs) and the Fast Fourier
curve technique. Jun and Gadala [4] present the analysis of
Transform (FFT) methods. Results show that the proposed
methodology can be applied to sound signal analysis; there this
dynamic behavior of a cracked rotor. Patel and Darpe [5]
detection technique is suited for detection of fault frequencies in present a vibration signature analysis of a rotor with a rotor-
induction motors. stator rub, transverse fatigue crack and unbalance. Yan and
Gao [6] present a filter construction technique for enhanced
Keywords—induction motor; fault detection; sound; intrinsic defect identification in rotary machine systems. Liu et al. [7]
mode function; spectral analysis. present a diagnostic scheme for bearing fault diagnostics using
neurofuzzy classifier.
I. INTRODUCTION Ever since the first motor was built, plant personnel have
Induction motors are widely used, and they are considered listened to the noises emanating from machines; with enough
critical components for electric utilities and process industries. experience, a listener may make a fairly accurate estimate of
An induction motor failure may yield an unexpected the condition of a machine. At best, this is merely a qualitative
interruption at the industry plant, with consequences in costs, analysis; at worst it may give an entirely wrong assessment.
product quality, and safety. These faults may be inherent to the Meanwhile very few works deal with sound signals for the
machine itself or to operating conditions. The origins of diagnosis and identification of faults on induction motor
inherent faults are due to the mechanical or electrical forces systems. Wu and Liao [9] present a neural network system for
acting on the machine. The greatest challenge in the area of automotive air-conditioner blower fault diagnosis using noise
condition monitoring is the diagnosis of a fault before it emission signal. Tinta et al. [10] present a technique that
becomes critical, and an early detection of this allows the repair provides a reliable estimate of fault diagnosis of vacuum
of the fault. In general, condition-monitoring schemes have cleaner motors. Wang et al. [11] present a numerical simulation
been widely used to sense specific failure modes in one of three for predicting the sound power from an inverter-driven
main induction motor components: the stator, the rotor, and the induction motor. The analysis of machine noise begins with an
bearings [1]. Traditionally, condition monitoring of induction understanding of the possible causes of the noise. In an
machines has been divided into two areas: mechanical induction motor, these include: slot harmonics, supply
problems and electrical problems [2]. Mechanical problems harmonics, rotor unbalance, winding asymmetry and the
include bearing wear, rotor unbalance, and airgap distortion. bearings. Some of these are inherent in the design of the
Electrical problems include broken rotor bars and winding machine and hence should remain constant over its lifetime

978-1-4673-6155-2/13/$31.00 © 2013 IEEE 149

(slot harmonics, winding asymmetry). Some develop over time, Where θ is the contact angle between the bearing surface,
and may lead ultimately to machine failure (bearing, unbalance, Dc is the cage diameter of the bearing and is measured from a
broken rotor bars). Others are due to external circumstances ball center to the opposite ball center, Db is the ball diameter
(supply harmonics). All of these add to the sound emitted by a
machine, resulting in a very complex audio signal. This work and N B is the number of balls in the bearing.
investigates a method that may be used to measure and analyze
the sound of an induction machine, where the feasibility of III. THEORETICAL BACKGROUND
distinguishing between broken rotor bars, bearing and
unbalance defects are the objectives of this investigation. A. Low-Pass Filter
The contribution of this investigation is the study for The sound signal is passed through a FIR filter with a
developing a condition monitoring strategy than can make a Kaiser window, as shown in Fig. 3. The filter has an order of
reliable assessment of the presence of a specific fault condition 128.
in an induction motor with a single fault present through the
analysis of sound signal. The sound signal analysis has shown B. Intrinsic Mode Function
good results in the identification of bearing and unbalance Intrinsic mode function (IMF) is a kind of signal that meets
faults in an induction motor. The proposed method uses a the physical interpretation of a single component signal. Huang
spectral analysis based on the IMFs and FFT algorithm, which et al. [14] have defined Intrinsic Mode Functions (IMFs) as a
are applied to the sound signal produced by an induction motor class of functions that satisfy two conditions:
for identification of the frequency-related fault. In this research,
two different faults in an induction motor such as bearing and • In the whole data set, the number of extrema and the
unbalance are investigated in an experimental way. Results number of zero-crossings must be either equal or differ
show the potentiality of the methodology as a deterministic at most by one.
detection technique that is suited for the detection of bearing
and unbalance faults in induction motors. • At any point, the mean value of the envelope defined
by the local maxima, and the envelope defined by the
local minima is zero.
II. FAULT-RELATED COMPONENTS
The next section explains the process to obtain IMFs called
Two different induction motor faults are considered in the Empirical Mode Decomposition.
paper: unbalance and bearing.
C. Empirical Mode Decomposition
A. Unbalance
To extract IMFs from the signal x ( t ) , a sifting process
The number of poles determines the speed of an induction
motor, and the speed of the motor can be identified by a peak in comprises the following steps:
the spectrum and then monitored at changes in amplitude. A Find the position and amplitudes of local maxima, and local
properly balanced and aligned motor has a frequency peak minima of x ( t ) . Then create an upper envelope by cubic spline
related to its speed that is barely visible. If a motor is out of
balance or misaligned, the signature of unbalance in a vibration interpolation of the local maxima, and a lower envelope by
signal, normally has the form of an increased amplitude along cubic spline interpolation of the local minima. Calculate the
the rotating frequency and its harmonics [12]. Vibration mean m1 ( t ) of the upper and lower envelopes. Subtracting the
analysis can provide a quick and relatively easy way to extract envelope mean signal from the original input signal, we have
information that may relate the unbalance fault in an induction
motor. h1 ( t ) = x ( t ) − m1 ( t ) (2)

B. Bearings Check whether h1 ( t ) meets the requirements to be an IMF.

Mc Fadden and Smith [13] give a review of the causes and If not, treat h1 ( t ) as new data and repeat the previous process.
expected frequencies of vibration due to rolling element Then set
bearings. From the geometry of the bearing, various theoretical
frequencies can be calculated such as the inner and outer race h11 ( t ) = h1 ( t ) − m11 ( t ) (3)
element pass frequencies, cage rotational frequency and rolling
element spin frequency. A defect on the outer race will cause Repeat this sifting procedure k times until h1k ( t ) is an IMF,
an impulse each time a rolling elements contact the defect. The this is designated as the first IMF.
rotor speed ( f r ) is the frequency at which the inner raceway
c1 ( t ) = h1k ( t ) (4)
rotates, that must be the frequency of the shaft. The theoretical
vibration frequency of the ball pass outer raceway frequency
Subtract c1 ( t ) from the input signal and define the
( f BPOF ) is easily determined as:
remainder, r1 ( t ) , as the first residue. Since the residue, r1 ( t ) ,
N B ⎛ Db ⎞ still contains information related to longer period components,
f BPOF = f r ⎜1 − cos θ ⎟ (1) it is taken as a new data stream. Repeat the above described
2 ⎝ Dc ⎠ sifting process to find more IMFs until the stopping criteria are

978-1-4673-6155-2/13/$31.00 © 2013 IEEE 150

met. The sifting process is stopped when either of criteria are between different surfaces of the motor components cause
met: 1) the component cn ( t ) , or the residue rn ( t ) , becomes so mechanical noise. The vibration signal arises from unbalanced
small in magnitude as to be considered inconsequential, or 2) rotating parts, where these vibrations increase at higher
rotational speeds or when a fault occurs in the induction motor.
the residue, rn ( t ) , becomes a monotonic function from which
The vibrational waves propagate all over the housing, thus
an IMF cannot be extracted. Finally, the signal can be creating reverberant vibrational fields that radiate sound. A
represented as the sum of IMFs and a residue. fault-free motor is significantly less noisy than a faulty motor,
n which increase substantially the intensity of mechanical noise
x ( t ) = ∑ c j ( t ) + rn ( t ) (5) in the lower range of rotational speeds and reach much higher
j =1 values. It is expected to find frequencies and harmonics in the
spectral analysis of sound signals related to the different
Fig. 1 shows five IMFs obtained from the sound signal. vibration faults in an induction motor.

IV. EXPERIMENTAL SETUP

Sound signal is captured in order to identify the dynamic
characteristics of the induction motor; and, simultaneously,
vibration signals are also acquired in order to verify the
obtained results. Forty tests were made for sound on each
condition in order to validate the propose methodology. Fig. 2
shows the experimental setup where different three-phase
induction motors (model WEG 00136APE48T, 740 Watts) are
used to test the performance of the proposed methodology
identifying the fault conditions treated in this work. The tested
motors have 2 poles, 28 bars and receive a power supply of 220
VAC at 60 Hz, and the applied mechanical load is that of an
ordinary alternator. The audio signal is acquired using a
condenser microphone JST model CX-509, which has a
Figure 1. IMFs of the sound signal.
cardioids polarization pattern. The microphone was placed in a
convenient position where the motor vibration did not have any
influence in the microphone. However, for different
D. Fast Fourier Transform microphone positions; the obtained results show some degree
Fourier analysis is very useful for many applications. To of variation. The audio signal is amplified utilizing a power
diagnose the fault, the frequency spectrum is obtained using the amp Marshall model MG15CDR. A 12-bit 4-channel serial-
Fast Fourier Transform (FFT) that is performed on the signal output analog to digital converter (ADC) (ADS7841) is used
under analysis. for data acquisition of audio. The instrumentation system uses a
sampling frequency of 4 kHz for obtaining samples of audio
The FFT is an algorithm that can efficiently calculate the during the induction motor steady state. Fig. 3 shows the
Discrete Fourier Transform (DFT). Let x0 ,..., xN −1 be the time proposed multiple fault diagnosis system for induction motor
series. The DFT is defined by the formula. monitoring utilizing audio signals.

N −1 n
− i 2π k
X (k ) = ∑ x (n) e N
(6)
n =0

E. Sound Analysis
Sound signals recorded in a noisy industrial environment by
using partial sound protection can still be utilized for feature
extraction. Namely, the anomalies in the spectrum of the sound
signals caused by the motor fault do not overlap with those
caused by environmental noise [15]. Li and Mechefske [16]
present the Wigner-Ville distribution technique applied for
identification of broken rotor bar and a combination of bearing
faults using current, vibration and acoustic methods. Various Figure 2. Test bench used during the experiment.
faults can be recognized from the sound signal in harmonics
related to the faults frequencies, for example, damages in Two different fault conditions are treated in this work:
bearing, unbalance and broken rotor bars. Tinta et al. [10] have
shown the study of a faulty motor, where the power-spectral-
A. Bearing Fault
density (PSD) at certain characteristic frequencies strongly
increase. Vibration of solid structures and mechanical contact To carry out the faulty bearing test, the bearing is
artificially damaged by drilling two holes with 1.191 mm of

978-1-4673-6155-2/13/$31.00 © 2013 IEEE 151

diameter on its outer race using a tungsten drill bit. Fig. 4a defect is found ( f BPOF = 177.7 Hz ) as is described in the section
shows the artificially damaged bearing 6203-2ZNR used in this
IV. The bearing fault is identified in the IMF 3, with the
work. The vibration characteristic defect frequency of the
presence of third-order harmonic of the ball pass outer raceway
rolling element bearing outer race is calculated by using
equation (1). The tested induction motor has a rotor frequency frequency ( f BPOF ) at 533 Hz in the spectrum obtained with
f r = 57.75 Hz and a test bearing having eight balls of diameter IMF and FFT, as shown in Fig. 5.
6.5 mm and the cage diameter of the bearing of 28 mm with
contact angle θ = 0 , thus, the ball pass outer raceway
frequency defect ( f BPOF ) is found to be 177.7 Hz.

Figure 3. Block diagram of the proposed fault diagnosis system

for induction motors.

B. Unbalance Fault
The unbalance condition is present when the induction
motor mechanical load is not uniformly distributed, taking the
center of mass of the motor shaft. Fig 4b shows a pulley with
an added mass used for generating unbalance on the induction
motor. The signature of unbalance in a vibration signal
normally has the form of increased amplitude along the rotor
frequency, being located in this case at 57.75 Hz.

Figure 5. Analysis region of sound signal for (a) a healthy motor and
(b) a motor with bearing fault.

B. Analysis of unbalance fault

The unbalance defect is verified with the increasing value
of the rotor frequency f r = 57.75 Hz in the vibration spectrum
as described in section IV. This fault is identified in the IMF 2,
Figure 4. Artificially generated faults (a) Outer race damaged bearing, with the presence of fourth-order harmonic located at 229 Hz in
(b) Unbalance pulley.
the spectrum obtained with IMF and FFT, as shown in Fig. 6.
Table 1 shows the detectability comparison in dB for the
V. EXPERIMENT RESULTS proposed methodology against the traditional FFT. The
The proposed sound analysis has been applied to several detectability is calculated as the amplitude ratio (in dB)
cases in which a single fault (bearing defect and unbalance) is between the faulty over the healthy condition.
present in the induction motor. The analysis of sound signal is
implemented in the Matlab Digital Signal Processing Toolbox.
TABLE I. DETECTABILITY IN DECIBELS FOR THE FAULT ANALYSIS

A. Analysis of bearing fault Condition FFT IMF & FFT

The detection of bearing defect is first identified in Bearing defect 0 9
vibrations signal where the ball pass outer raceway frequency Unbalance 8 19

978-1-4673-6155-2/13/$31.00 © 2013 IEEE 152

 REFERENCES

[1] Benbouzid M. E. H., "A review of induction motors signature analysis as

a medium for faults detection" IEEE Transactions on Industrial
Electronics, vol. 47, pp. 984-993, October 2000.
[2] Al Kazzaz S. A. S. and Singh G. K., "Experimental investigations on
induction machine condition monitoring and fault diagnosis using digital
signal processing techniques" Elsevier Electric Power Systems Research,
vol. 65, pp. 197-221, June 2003.
[3] Babu T. R. and Sekhar A. S., "Detection of two cracks in a rotor-bearing
system using amplitude deviation curve" Elsevier Journal of Sound and
Vibration, vol. 314, pp. 457-464, July 2008.
[4] Jun O. S. and Gadala M. S., "Dynamic behavior analysis of cracked
rotor" Elsevier Journal of Sound and Vibration, vol. 309, pp. 210-245,
January 2008.
[5] Patel T. H. and Darpe A. K., "Coupled bending-torsional vibration
analysis of rotor with rub and crack" Elsevier Journal of Sound and
Vibration, vol. 326, pp. 740-752, October 2009.
[6] Yan R. and Gao R., "Energy-based feature extraction for defect
diagnosis in rotary machines" IEEE Transactions on Instrumentation
Measurement, vol. 58, pp. 3130-3139, 2009.
[7] Liu J., Wang W. and Golnaraghi F., "An enhanced diagnostic scheme for
bearing condition monitoring" IEEE Transactions on Instrumentation
and Measurement, vol. 59, pp. 309-321, February2010.
[8] Rodriguez-Donate, C., Romero-Troncoso, R. J., Garcia-Perez, A. and
Razo-Montes D. A., "FPGA based embedded system for induction motor
failure monitoring at the start-up transient vibrations with wavelets"
Proceedings of the IEEE 3rd. International Symposium on Industrial
Embedded Systems SIES 2008, Montpellier, France, pp. 232-238.
[9] Wu J. and Liao S., "Fault diagnosis of an automotive air-conditioner
blower using noise emission signal" Elsevier Expert Systems with
Applications, vol. 37, pp. 1438-1445, March 2010.
[10] Tinta D. et al., "Fault diagnosis of vacuum cleaner motors" Elsevier
Control Engineering Practice, vol. 13, pp. 177-187, February 2005.
[11] Wang C., Astfalck A. and Lai J. C. S., "Sound power radiated from an
inverter-driven induction motor: experimental investigation" IEEE
Figure 6. Analysis region of sound signal for (a) a healthy motor and Electric Power Applications, vol. 149, pp. 46-52, January 2002.
(b) a motor with unbalance fault.
[12] Bloch H. P., "Vibration analysis in machinery failure analysis and
troubleshooting" Houston: Gulf Professional Publishing, 2009.
VI. CONCLUSION [13] McFadden P. D. and Smith J. D., "The vibration produced by multiple
points defects in a rolling element bearing" Elsevier Journal of Sound
In this paper, a new methodology for detecting single faults and Vibration, vol. 98, pp. 263-273, January 1985.
in induction motors has been proposed. It has been shown that [14] Norden E. Huang et al., "The empirical mode decomposition and the
the power-spectral-density of sound signals for bearing and Hilbert spectrum for nonlinear and non-stationary time series analysis"
unbalance faults exhibit harmonic frequencies, related to the Proceedings of the Royal Society of London Series a-Mathematical
Physical and Engineering Sciences, vol. 454, pp. 903-995, March 1998.
different faults in an induction motor. However, for further
[15] Benko U., Petrovcic J., Juricic D., Tavcar J., Rejec J. and Stefanovska
development, there is still the necessity of making more A., "Fault diagnosis of a vacuum cleaner motor by means of sound
investigation to find the best microphone position, or to use analysis" Elsevier Journal of Sound and Vibration, vol. 276, pp. 781-
several microphones placed in different positions and use 806, September 2004.
special algorithms for noise cancellation, can be considered as [16] Li E. and Mechefske C., "Detection of induction motor faults: a
an option for future investigations. The proposed method in this comparison of stator current, vibration and acoustic methods" Journal of
paper can be extended for further development and Vibration and Control, vol. 12, pp. 165-188, 2006.
implementation into a field-programmable gate array system [17] Rodriguez-Donate, C., Romero-Troncoso, R. J., Cabal-Yepez, E.,
Osornio-Rios, R. A. and Garcia-Perez, A., "Wavelet-based general
for online applications of surveillance and diagnostics of the methodology for multiple fault detection on induction motors at the
reported and other faults present in induction motors. startup vibration transient" Journal of Vibration and Control SAGE, vol.
17, pp. 1299-1309, 2011.

978-1-4673-6155-2/13/$31.00 © 2013 IEEE 153