Rotor Fault Detection in Line-fed Induction Machines Using

Complex Wavelet Analysis of Startup Transients 325

15
x

Rotor Fault Detection in Line-fed Induction

Machines Using Complex Wavelet
Analysis of Startup Transients
Fernando Briz, Michael W. Degner†, Pablo García,
David Díaz-Reigosa and Alberto Díez
University of Oviedo, †Ford Motor Company
Spain, †USA

1. Introduction
Three–phase inductions machines are the primary consumers of electric power in
industrialized countries, they can typically consume between 40 to 50% of all the generated
power (Thomsom et al., 2001). Diagnostic testing of induction machines is therefore of
tremendous importance in many applications, and has been the focus of intense research for
many years (Aller et al., 2002), (Antonino-Daviu et al., 2006), (Briz et al., 2008), (Benbouzid,
2000), (Douglas et al. 2003), (Douglas et al. 2004), (Douglas et al. 2005), (Douglas et al. 2005-b),
(Faliang Niu & Jin Huang, 2005), (Nandi & Toliyat, 2005), (Thomson & Fenger, 2001),
(Supangat et al., 2006).
Based on the results from failure surveys, the failure mechanism for induction machines can
be divided into the following four general categories, with the percentage of total failure
shown (Thomsom et al., 2001).
 Stator related: 40%
 Rotor related: 10%
 Bearing related: 40%
 Other: 10%

A number of methods have been developed for the purpose of detecting faults in induction
machines, the ultimate goal being to prevent unexpected equipment downtime or severe
equipment damage. Classification of the existing methods can be done according to several
criteria, including:
 The type of fault that can be detected (stator, rotor, bearing, … ).
 The physical quantities that are measured and further processed for the detection
(phase currents, voltages, temperature, infrared, vibration, axial flux, acoustic
noise, …).

This work was supported in part by the Research, Technological Development and Innovation Program
of the Spanish Ministry of Science and Education-ERDF under grant MEC-ENE2007-67842-C03-01.

www.intechopen.com
326 Fault Detection

 If they work on-line or off-line. On-line methods do not stop the normal operation
of the machine, while off-line methods require discontinuing the normal operation
of the machine and often require moving it to a test rig.
 If they can work with constant and/or variable fundamental excitation, i.e., if they
are valid for line-connected machines and/or inverter fed machines.

On-line methods are preferred in general for obvious reasons. Of special interest are
methods that use terminal electrical quantities (currents/voltages), as they are easily
measured. Methods of this type include negative sequence currents/impedances and motor
current signature analysis (MCSA) (Kholer et al., 2002), (Lee et al., 2003), (Nandi & Toliyat,
2005), (Thomson & Fenger, 2001). These methods exploit the fact that an imbalanced
machine, i.e., faulted, when fed with a balanced, three-phase voltage, produces specific
components in the stator currents whose magnitudes and frequencies depend on the level of
asymmetry and the cause of the fault condition.
Independent of their attractive properties, these methods have drawbacks that can limit
their performance (Kholer et al., 2002), (Lee et al., 2003), (Nandi & Toliyat, 2005), (Thomson
& Fenger, 2001). First, the line voltages feeding the machine often contain negative
sequence components, which produce spurious current components caused not by
imbalances in the machine but by the voltage feeding it. Although it is possible to measure
the level of imbalance in the line voltages, it is not easy to accurately estimate their
contribution to the overall current, since it depends on the machine’s parameters and
operating point. A second limitation of these techniques is the dependence of the fault
related components on the operating point of the machine, and specifically, on the load level
or slip. Several improvements have been proposed to mitigate these drawbacks (Lee et al.,
2003), (Nandi & Toliyat, 2005).

1.1 Rotor fault detection

Rotor faults account for a relatively small portion (10%) of the overall failures occurring in
induction machines. However, this portion is not uniform for all machines designs and
applications. Broken rotor bars primarily occur in medium voltage motors with copper bar
rotors. Direct line starting with heavy loads causes high thermal and mechanical stresses;
pulsating mechanical loads such as reciprocating compressors or coal crushers (etc.) can
subject the rotor cage to high mechanical stresses; an increased risk of rotor failure can exist
in these cases (Loránd et al., 2004).
MCSA methods have successfully been applied to rotor fault detection. Damaged rotors
produce spectral components in the stator current at frequencies that are function of the slip
(and therefore of the rotor speed). The magnitude of these components is a function of the
level of asymmetry of the rotor, from which the rotor condition can be evaluated.
Regardless of its simple physical principles, use of MCSA for rotor fault detection presents
limitations. When the machine operates with light load (small slips), the magnitude of the
fault related components decrease, and their frequencies gets close to large fundamental
excitation frequency related components, which result in them being difficult to separate.
Also, oscillating loads or imbalances in coupled gears and/or mechanical transmissions can
produce components in the currents at frequencies similar to those caused by faults, making
them difficult to distinguish from a real fault.

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 327

The startup transient offers opportunities for performing rotor diagnostics of line-connected
machines. When a machine is connected to the line, the startup transient is characterized by
large stator (and rotor) currents, as well as by large slips (i.e., rotor speed significantly
smaller than the excitation frequency). Damaged rotors create large rotor speed dependent
components in the stator current during the startup transient. Analysis of the stator current
of line-connected machines during startup provides, therefore, an excellent opportunity for
performing rotor diagnostics.
Detection of damaged rotor bars using startup transients requires the use of methods
capable of processing the transient signals. Wavelet functions are well suited for this
purpose and wavelet based analysis has been proposed for machine diagnostics using
startup transient currents (Aller et al., 2002), (Antonino-Daviu et al., 2006), (Briz et al., 2008),
(Douglas et al. 2003), (Douglas et al. 2004), (Douglas et al. 2005), (Douglas et al. 2005-b),
(Faliang Niu & Jin Huang, 2005), (Nandi & Toliyat, 2005), (Thomson & Fenger, 2001),
(Supangat et al., 2006). In most of these works, standard wavelets transforms, i.e., wavelet
families proven useful in other applications, are used. Discrete wavelet transform based
filtering was used in (Aller et al., 2002), Daubechies-8 wavelet in (Douglas et al. 2003),
(Douglas et al. 2004), (Douglas et al. 2005), (Douglas et al. 2005-b), (Supangat et al., 2006),
Daubechies-40 wavelet in (Antonino-Daviu et al., 2006). In (Faliang Niu & Jin Huang, 2005),
a continuous Morlet wavelet is proposed and the analysis was based on an estimate of the
torque. This requires measurement of the stator voltage, as well as an estimate of the stator
resistance, which significantly complicates the implementation of the method. A significant
inconvenience of standard wavelets is that no clear criteria for selecting the wavelet function
exist, with selection being made ad hoc. Also, although the wavelet based analysis can reveal
differences between a healthy and a damaged machine, there is not a clear relationship
between the physical characteristics of the signal and the results of wavelet based analysis.
Alternatively to the use of standard wavelets, it is possible to design wavelet transforms
based on the physical properties exhibited by the currents when a damaged rotor exist (Briz
et al., 2008). Since these wavelet respond to a known pattern, the design, as well as the
selection of the parameters, are deterministic, with the interpretation of the results of the
wavelet transform being straightforward.
A method for rotor fault detection using wavelet based analysis of the stator current
complex vector during startup is presented in this work. The chapter is organized in 8
sections. Section 2 introduces the stator current complex vector of three-phase machines,
with its nature during start-up transients for both healthy and faulty machines being
discussed in section 3. Section 4 introduces complex vector wavelets designed to detect
rotor fault related patterns of the stator current complex vector during the startup transient.
Discussion on the design of the wavelet function, as well as experimental results
demonstrating the viability and performance of the method are presented in Section 5.
Section 6 analyses implementation aspects, the conclusions being presented in Section 7.
Section 8 provides the reference list.

www.intechopen.com
328 Fault Detection

phase b s s
ib ids iqds
va ia s
ic iqds


vb ib

phase a s
vc ic phaseq
ia s
iqs

phased
s
phase c

a) Three phase machine b) Three phase system c) Equivalent two-phase

system
Fig. 1. Three phase machine and reference coordinates systems.

Table 1. Induction motor parameters (fe = 50 Hz)

Power Rating 0.9 kW, 4 poles
V rated/ I rated (wye) 400 V / 2.3 A (rms)
Rated speed 1425 rpm
# Stator Slots / #Rotor 24 / 30
Slots

2. The stator current complex vector

The stator current complex vector is the quantity measured and used in this work for
detecting rotor faults in induction machines. This section provides a basic background on
the definition and utilization of complex vector variables for the analysis of ac electric
machines.
Complex vector quantities (or alternatively qd-axis models) are widely used for the analysis,
design, control and diagnosis of three-phase, ac machines and three-phase power systems
(Novotny & Lipo, 1996). Given a set of three-phase currents ia, ib and ic (Fig. 1a), the stator
current complex vector is defined as:

s s s 2 j2/3 j4/3
i qds = i qs – j i qd = ( ia + ib e + ic e ) (1)
3

The transformation in (1) can be visualized as the currents ia, ib and ic aligned with three non-
orthogonal axes with a phase shift of 120º from each other (Fig. 1b), being transformed to an
orthogonal q-d reference frame (Fig. 1c). The stator current complex vector (1) can be
represented as a complex quantity or in cartesian form by using its real and imaginary
s s
components, iqs and ids respectively.
Fig. 2a shows the three phase stator currents of the test machine operating in the steady-
state. The parameters of the machine are shown in Table I. Fig. 2b shows the resulting q-
and d-axis components obtained using the transformation defined by (1), while Fig. 2c

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 329

shows the trajectory of the resulting complex vector. For the case of balanced, sinusoidal
three-phase currents, the stator current complex vector consists of a single rotating component,
with its magnitude being equal to the peak value of the phase currents, its frequency of
rotation coinciding with the frequency of the phase currents.

a)
2
ia, ib, ic
(A) 0

-2

b)
s 2
Real ( iqds ),
s 0
Imag ( iqds )
(A) -2

0 40 80
time (ms)
c)

s
iqds
(A)
80
time (ms)

40

0 2
-2 0
0
s
s 2 -2 Imag ( iqds )
Real ( iqds )
Fig. 2. a) phase current iu, iv, iw, b) q and d-axis components of the stator current complex
vector, and c) trajectory of the stator current complex vector with the machine operated in
steady-state.

s 1
|fft( iqds )|
0.1
(A)
0.01
0.001

-400 -200 0 200 40

frequency (Hz)
Fig. 3. Frequency spectrum (only magnitude) of the stator current complex vector.

www.intechopen.com
330 Fault Detection

a)
20
s
| iqds | 15
10
(A)
5
0
b)
50
r 40
30
(Hz)
20
10
0
0 10 20 30
time (s)
Fig. 4. a) Magnitude of the stator current complex vector and b) rotor speed during startup
(400 V rms, 50 Hz), with the motor driving a large inertia load.

Fig. 3 shows the frequency spectrum of the stator current complex vector. Since it is a
complex vector quantity, its frequency spectrum contains both positive (forward rotating)
and negative (backward rotating) frequency components. It can be observed from the figure
that the largest component is at a frequency of e=50 Hz, which corresponds to the
fundamental (line) frequency, with a number of additional harmonics of relatively small
magnitude being also observable both at positive and negative frequencies.

3. Analysis of the stator current complex vector during startup

Fig. 4 shows the stator current complex vector magnitude and the rotor speed of the test
machine after it was connected to the line. Since the transient stator current complex vector
after connection is a non-stationary signal, conventional FFT based analysis of the type
shown in Fig. 3 is not adequate. Instead, the Short Time Fourier Transform (STFT) can be
used for these purposes (Benbouzid, 2000), (Briz et al., 2008). Fig. 5a shows the spectrogram
of the stator current complex vector (1), obtained using the STFT, during the transient shown
in Fig. 4, for the case of a healthy machine. A sampling frequency of 5 kHz, with a window
width of 2048 samples was used.
The most significant component in the spectrogram is at the line frequency, e=50 Hz, and
corresponds to the fundamental current. Additional components of reduced magnitude
(note the logarithmic scale of the color bar) can also be observed. Horizontal lines
correspond to components that do not vary with the rotor speed and have several physical
origins, e.g., saturation; additional components in the line voltages; interactions between the
fundamental excitation and asymmetries intrinsic to the design of the machine; imbalances
in the measurement circuits (current sensors and further electronics), etc. All of these
components generally have a small magnitude compared to the signals of interest but they
are still visible when a logarithmic scale is used (Briz et al., 2008).

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 331

10

400

1
a) Healthy machine 200
frequency (Hz)
0
0.1

-200 er
0.0
-400

0 10 20 30 40

10

400
b) Machine with a broken rotor bar

1
200
frequency (Hz)

0
0.1

-200 er
0.0
-400

0 10 20 30 40
time (s)
Fig. 5. Spectrogram of the stator current complex vector during startup with the motor
driving a large inertia load, for the case of a) a healthy machine and b) machine with a
broken rotor bar. Color bar units: Amps.

Rotor speed dependent components can also be observed in the spectrogram. Such
components have previously been reported even for the case of healthy machines due to
design and construction asymmetries (Benbouzid, 2000).
Fig. 5b shows the spectrogram of the stator current complex vector during a startup
transient for the case of a machine with a broken rotor bar. The bar was disconnected by
drilling the end ring, as shown in Fig. 6. The rotor was modified so that the continuity of the
end-ring was maintained and the rotor laminations were not affected. Components of the
stator current complex vector caused by a damaged rotor bar are given by (2) (Benbouzid,
2000), where e is the electrical frequency, s is the slip, p the number of pole pairs,

www.intechopen.com
332 Fault Detection

Fig. 6. Rotor modified to break a rotor bar by drilling the end-ring.

and with k/p = 1, 5, 7, 11, 13, … . The most important component, (3), denoted as brb_2,
is obtained from (2) with k/p = 1.
 (1  s ) 
brb  e  k  s  (2)
 p 

brb_2 = e (12s)= e + 2 r (3)

Using (2) and (3), the stator current complex vector can be modeled as (4), where the first
term on the right hand side represents the current of a healthy machine, while the rest of the
terms represent the rotor fault-induced components.

e j (   e  2 r ) e j ((   e 2 r ) k/p )
s
iqds s
= iqds healthy + Ibrb_2 + Ibrb_k (4)

It can be seen from Fig. 5b that rotor speed dependent components that are readily
observable during the startup transient have a significantly reduced magnitude in the
steady-state. In addition, all of the rotor speed dependent components converge to
frequencies in the steady-state that are spectrally close to the fundamental excitation frequency
or its harmonics. This makes their separation in the steady-state difficult, especially if the
machine is not heavily loaded. It can be concluded then that significantly richer information
for revealing a damaged rotor exists in the current vector during startup than exists during
steady-state operation.

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 333

4. Detection of damaged rotor bards using complex vector wavelets

The separation of the information revealing a damaged rotor from the overall stator current
requires adequate signal processing. The STFT, combined with a spectrogram
representation of Fig. 5 is one option. However, while visually insightful, it is not easy to
establish a metric that indicates the rotor condition. Wavelets are an efficient tool for
pattern detection in transient signals. This can be accomplished by designing the wavelet to
resemble the attributes of the signal to be detected.

4.1 Wavelet design

It has been shown that a damaged rotor induces, among others, a component, Ibrb_2 (4), in
the stator current complex vector. This component is resembled during a startup transient
by a wavelet function , (5), consisting of a complex exponent multiplied by a windowing
function, h.

j(( tqe  t   2qr  t  )

(t) = h(t) e (5)

The complex exponent is a function of the electrical angle, e and the rotor angle, r. These
angles can be obtained by integration of the fundamental excitation frequency, e, and the
rotor speed, r, respectively, (6). However, while e is known in advance and constant, r
is practically never measured in line-connected machines. Because of this, an estimate of the
rotor speed during the startup transient rw needs to be used. For this work, the estimate
rw(t) was obtained using a simple polynomial curve fit method from the e+2r
component in the spectrogram shown in Fig. 5a during a commissioning stage, the
estimated rotor speed being shown in Fig. 4b. The angle rw was then obtained by
integrating rw(t) (6).

e(t) =
 e 
dt , rw(t) = rw (t ) dt (6)

It is noted that the determination of the exact value of rw(t) is not critical, as the wavelet (5)
will later be scaled (i.e., stretched or shrunk) to effectively sweep the expected range of
startup transient time lengths and shifted to find the exact time in which the startup
transient occurred. This is done by defining rw (7) and h as a function of scale (denoted as
a) and time translation (denoted as b) (discussion about the design of h is presented in sub-
section 4.3). The resulting wavelet function is given by (8).

rw(a,b) =
 rw ((t  b ) / a ) dt (7)

j ( tqe  2qrw a , b  )
(a,b) = h(a,b) e (8)

Since the wavelet definition (8) relies on an estimation of the rotor speed shape, the reliability
of the method can be reduced if significant variations in the rotor speed shape, due to variations

www.intechopen.com
334 Fault Detection

0.6

0.4 broken rotor bar

(A)
0.2
health
0
0 10 20 30 40
r (Hz)
Fig. 7. Magnitude of the stator current complex vector component at e+2r (Ibrb_2 in (4))
vs. r for the case of a healthy machine and a machine with a broken rotor bar, during the
startup transients of Fig. 5.

in the load characteristics, occur. It is emphasized however that variations of the transient
length do not prevent the method from providing reliable fault detection, as will be
demonstrated in section 5.

4.2 Wavelet transform

Once  is defined, (8), the coefficient of the wavelet transform C is obtained using (9)
(MathWorks Inc., 2007), with ‘*’ standing for the complex conjugate. It is noted that,
although the integral symbol is used in (9), it actually operates with sampled signals, with
the integration being transformed into a summation in the practical implementation.

  ( a, b) i
1 * s
C(a,b) = qds dt (9)
a

This coefficient is a function of a, i.e., how much the base wavelet is stretched or shrunk, and
b, i.e., how much the wavelet is shifted in time with respect to the signal being analyzed.
Evaluation of (9) is made by changing a and b at short, regular steps, resulting therefore in a
continuous wavelet transform (MathWorks Inc., 2007).

Selecting a and b in (9) is straightforward since they are directly related to characteristics of
the signal being analyzed. The limits for a are related to the minimum and maximum
startup transient time lengths, while the limits for b are related to how accurately the startup
transient can be detected.

4.3 Selection of the windowing function

The windowing function h in (5) is required so that the wavelet, , has a finite length and a
smooth transition, since the complex exponential term in (5) has a constant magnitude equal
to one. Fig. 7 shows the magnitude of the stator current complex vector e2r component
vs. r for the case of a healthy machine and for the case of a machine with a broken rotor bar.
They were obtained from signal processing of the spectrograms in Fig. 5a and 5b,
respectively. It is interesting to notice that while the differences between the two signals are
noticeable during the startup transient, both signals increase significantly and coincide for
values of r near e (50 Hz). This is due to the fact that the frequency being tracked (3)

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 335

approaches the fundamental frequency of the current, which has a much larger magnitude

50
a)
rw

(Hz)
25
(Hz)

b) 1
h ||
magnitude

0
t1 t2 t3 t4
1
c)
real and imag

Real(),
Imag() 0

-1

180
d)
,
angle

(deg.) 0

-180

0.1 0.2 0.3 0.4 0.5 0.6

time (s)
Fig. 8. Construction of the complex vector wavelet, . a) estimated rotor speed, b)
windowing function, h, c) real and imaginary components of , and d) phase angle of .

and is difficult to separate from the desired signal. A similar effect is also observed for
values of r near 0, caused by the spectral closeness of the e component.

To avoid border effects, h is chosen to be equal to zero for values of rw near 0 and e. This is
schematically shown in Fig. 8. The window has two transition regions (t1t2 and t3t4) of
variable magnitude and a mid-region (t2t3) of constant magnitude. The transition regions
are defined by (10)-(11), being similar to a hanning window. The real and imaginary
components of the resulting wavelet function, (5), are shown in Fig. 8c, with its magnitude
and phase in Fig. 8b and 8d, respectively.

1  cos(12 ) (t  t1 )
h12  , with 12  and t1  t  t2
( t 2  t1 )
(10)
2

www.intechopen.com
336 Fault Detection

1  cos( 34 ) (t  t3 )
h34  , with  34  and t3  t  t 4
(t 4  t 3 )
(11)
2

50 scale=0.2
a)
rw(t/a) scale=1
(Hz) 25
scale=2

scale=0.2
b) 1
h(t/a)
scale=1 scale=2
0.5

0
c)
s 20 scale=0.2
|h· iqds | scale=1
15
|· iqds |
s 10

(A)
5 scale=2
0
0 10 20 30 40
time (s)
Fig. 9. a) Estimated rotor speed, b) windowing function and c) magnitude of windowed
stator current complex vector resulting from applying the windowing functions to the stator
current complex vector magnitude of Fig. 1. In all cases b=0.

Although this window will be used for all of the experimental results presented in this work, a
variety of windowing functions can be used, each providing similar results (Briz et al., 2008).

5. Experimental verification of the method

The proposed method was tested using the machine in Table I. Fig. 9a and 9b shows the
estimated rotor speed rw and the windowing function, h, defined by (10)-(11), for different
scales of a, with 0.2 and 2 representing the limits over which the wavelet transform (9) was
evaluated, and 1 being a nominal case. Fig. 9c shows the magnitude of the windowed stator
current complex vector for these values of a.
Fig. 10a and 10b show the magnitude of the C coefficient resulting from the wavelet
transform for the case of a healthy machine and a machine with a broken rotor bar
respectively. The estimated start of the transient was assigned a time t=0, the b (time
translation) parameter of the wavelet transform, , being evaluated in a range of 0.2 to 0.3 s
around this value. The a parameter was varied within a range of 0.2 to 2 times the base
function. The region of Fig. 10b that shows the most significant differences with respect to
Fig. 10a is zoomed. Comparing Fig. 10a and 10b it can be observed that the coefficient of the

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 337

wavelet transform for the case of a healthy machine (Fig. 10a) has a rather constant value
(uniform color), independent of the values for a and b. This means that the wavelet transform
1.6

1.8 1.4

1.6
1.2
a) Healthy machine

1.4
1
1.2
a (scale)

0.8
1
0.6
0.8

0.4
0.6

0.4 0.2

0
-0.2 -0.1 0 0.1 0.2 0.3
1.6
0.88
1.8 1.4
0.86
b) Machine with a broken rotor bar

1.6
0.84 1.2

1.4
-0.1 0 0.1 1
1.2
a (scale)

0.8
1
0.6
0.8

0.4
0.6

0.4 0.2

0
-0.2 -0.1 0 0.1 0.2 0.3
b (time translation, seconds)
Fig. 10. Magnitude of the wavelet transform coefficient |C| during the startup transient
shown in Fig. 9, as a function of a and b (relative to the estimated start of the transient).

1.5
1 broken rotor bar
|C| 1

0.5
health
0
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
a
Fig. 11. Magnitude of the wavelet transform coefficient |C| during a startup transient, as a

www.intechopen.com
338 Fault Detection

function of a (scale), (b=0.04 s), for the case of a healthy machine and a machine with a
broken rotor bar.
s
| iqds |, 20
s 15
|h· iqds |,
10
(A) 5
|h iqds | | iqds |
0
0 2 4 6 8 10 12
time (s)
Fig. 12. Magnitude of the stator current complex vector and of the windowed stator current
complex vector during startup transient.
1.6

1.8 1.4

1.6
1.2
a) Healthy machine

1.4
1
1.2
a (scale)

0.8
1
0.6
0.8
0.4
0.6

0.4 0.2

0
-0.2 -0.1 0 0.1 0.2 0.3
1.6
0.31
1.8 1.4
0.29
b) Machine with a broken rotor bar

1.6
1.2
0.27
1.4
-0.1 0 0.1 1
1.2
0.8
a (scale)

1
0.6
0.8

0.4
0.6

0.4 0.2

0
-0.2 -0.1 0 0.1 0.2 0.3

b (time translation, seconds)

Fig. 13. Magnitude of the wavelet transform coefficient |C| during the startup transient
shown in Fig. 12, as a function of a and b (relative to the estimated start of the transient).

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 339

1.5

1 1 broken rotor bar

|C|
0.5
health
0
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
a
Fig. 14. Magnitude of the wavelet transform coefficient |C| during a startup transient, as a
function of a (scale), (b0.04 s), for the case of a healthy machine and a machine with a
broken rotor bar.

Wavelet design Measurement

Base rotor speed Windowing function of phase currents
rw h (startup transient)

Selection of Calculation
s
a & b limits of iqds (1)

scaling & time scaling & time

translation of rw translation of h

Calculation of the wavelet function  (5)

Calculation of the wavelet coefficients C (9)

Rotor fault detection (peak value of C)

Wavelet based analysis for each startup transient

Fig. 15. Flowchart showing the implementation of the method.

effectively eliminates components of the stator current complex vector different from the one
modeled by the wavelet function . On the other hand, the coefficient C resulting from the
wavelet transform for the case of a machine with a broken rotor bar has large values for well
defined values of a and b, which reflects that good correlation exists between the wavelet, ,
and the signal for these values of a and b.
Fig. 11 shows the value of C, as a function of a, from Fig. 10 for b=0.04 s, for the cases of a
healthy machine and a machine with a broken rotor bar. It can be observed that the
maximum value of C is obtained for a=0.86, i.e., the startup transient was shorter than the
wavelet base function.

www.intechopen.com
340 Fault Detection

It is concluded from these results that the proposed wavelet transform is highly sensitive to the
properties exhibited by the current of a motor with a damaged rotor during startup. A metric
as simple as the peak value of |C| can be used to effectively indicate the condition of the
machine.

5.1 Influence of the start up transient length

To assess the robustness of the method against variations in the startup transient
characteristics, the analysis was repeated with the machine driving a significantly smaller
inertia. Fig. 12 shows the magnitude of the stator current complex vector during the startup,
as well as the windowed current. Fig. 13a and 13b shows the magnitude of the coefficient,
C, for the case of a healthy machine and of a machine with a broken rotor bar, respectively.
Regardless of the significant change in the startup transient's length, the wavelet transform
in Fig. 13b clearly displays the presence of a fault. Fig. 14 shows the values of C, as a
function of a, for b= 0.06 s, both for the healthy machine and for the machine with a broken
rotor bar.

6. Implementation of the method

The proposed method requires the measurement of three current sensors for the
implementation described in Section 3 (see Fig. 1a). The use of two current sensors is
adequate for these purposes, however, and is a common practice, provided that no zero
sequence current exists. It is equally valid for both delta- and wye-connected machines, and
for machines with open as well as closed rotor slots. For the experiments presented in this
work, conventional Hall effects current sensors and 12 bits A/D converters were used.
Fig. 15 shows the flowchart of the method. For the wavelet design, the rotor speed, rw,
during a startup transient is estimated and stored. The windowing function, h, is then
designed as discussed in Section 5 and stored for later use. The stored rw and h are then used
for the wavelet based analysis each time that a startup occurs.
The peak value of |C|, is used to indicate the condition of the machine. This implies the
establishment of a threshold that differentiates a healthy from a faulty machine. This cannot
be easily done when no previous data from the machine of interest exists. It is
recommended that Cpeak be measured when a machine is first installed, or is known to be
healthy, after that, increments or changes in the value of Cpeak would indicate a
deterioration of the rotor. Tracking the changes in Cpeak could also be used for diagnostics
in machines already installed and for which the condition of the rotor is unknown (with the
assumption that the rotor starts out relatively healthy and any increase in C indicates a
deterioration in the rotor bar health).
The experimental results shown in Fig. 10 and 13 used a wide range of values for a and b to
better show the results of the method. However, to reduce the computational and time
requirements of the method, the range of evaluation for a and b parameters can be
significantly reduced by simple analysis of the startup transient current vector magnitude
(Fig. 4 and 12). The start of the transient can typically be accurately established, which
suggests that b could be restricted to a single value. However, it is convenient in practice to
allow a narrow range of evaluation for b. This allows for compensation of effects like those

www.intechopen.com
Rotor Fault Detection in Line-fed Induction Machines Using
Complex Wavelet Analysis of Startup Transients 341

caused by the non-ideal (non-repetitive) behavior of the breakers, the variation of the instant
in time relative to the period of the phase voltages in which the breakers were ordered to
close, as well as for inaccuracies in the rotor speed rw. For the experiments presented in
this paper, the time translation parameter b was changed in steps of 5 ms, with a range of
variation in the order of 0.1 s relative to the theoretical start of the transient.
As for the selection of the range of a, line-connected machines often show repetitive startup
transients, allowing for narrow bounds to a, which would be specific for each application.
For the case of machines showing significant variations in the startup transient length, simple
threshold based analysis of the startup current can be used to dynamically adapt the limits for
a for each startup transient. For the experiments presented in this paper, a was changed in
steps of 0.02. The calculation of the C coefficient for each value of a and b, including scaling
of h and rw, took 0.06 s in a standard computer.
In the top-right corner of Fig. 10b and 13b the wavelet transform coefficient is shown with a
and b restricted to a reduced range of values, which were automatically obtained by the
analysis of startup transient current described above.

7. Conclusion
Broken rotor bar detection in a line-connected induction machine using complex wavelets to
analyze the stator currents during startup transients was presented in this work. These
wavelets allow for the accurate detection of fault related components in the stator current
complex vector indicative of damaged rotors. In comparison with standard wavelet based
methods, the wavelet design is made following a well defined procedure. Because of this,
limits for the scale and time translation parameters of the transform are easily established,
reducing the computational requirements of the transformation. In addition, interpretation of
the results of the wavelet transform is physically insightful. Experimental results confirm the
effectiveness of the method in detecting damaged rotor bars.

8. References
Aller, J. M., Habetler, T. G. , Harley, R. G. , Tallam, R. M. , and Lee, S. B. ,“Sensorless Speed
Measurement of AC Machines using Analytic Wavelet Transform,” IEEE
Transactions on Industry Applications, vol. 38, pp. 1344–1350, Sept./Oct. 2002.
Antonino-Daviu, J.A., Riera-Guasp, M., Folch, J.R.; Palomares, M.P.M., “Validation of a new
method for the diagnosis of rotor bar failures via wavelet transform in industrial
induction machines”, IEEE Transactions on Industry Applications, vol. 42, no. 4,
pp:990–996, July-Aug. 2006.
Benbouzid, M.E.H., “A review of induction motors signature analysis as a medium for faults
detection”, IEEE Transactions on Industrial Electronics., vol. 47, no 5, Oct. 2000, pp
984-993.
Briz, F., Degner, M. W. , García, P., Bragado, D., “Broken Rotor Bar Detection in Line-fed
Induction Machines Using Complex Wavelet Analysis of Startup Transients”, IEEE
Transactions on Industry Applications, vol.: 44, nº 3, pp. 760-768, May. 2008.

www.intechopen.com
342 Fault Detection

Douglas, H.; Pillay, P.; Ziarani, A., “Detection of broken rotor bars in induction motors using
wavelet analysis”; IEEE International Electric Machines and Drives Conference
IEMDC'03, Vol. 2, pp: 923–928, June 2003.
Douglas, H.; Pillay, P.; Ziarani, A.K., “A new algorithm for transient motor current signature
analysis using wavelets”, IEEE Transactions on Industry Applications, vol 40, no. 5,
Sept. 2004, pp 1361-1368.
Douglas, H.; Pillay, P., “The Impact of Wavelet Selection on Transient Motor Current
Signature Analysis”, IEEE-International Conference on Electric Machines and Drives
IEMDC, 2005 pp 80-85.
Douglas, H., Pillay, P. and Ziarani, A. K. , “Broken Rotor Bar Detection in Induction
Machines with Transient Operating Speeds,” IEEE Transactions on Energy
Conversion, vol. 20, no. 1, pp. 135-141, March 2005.
Faliang Niu; Jin Huang, “Rotor broken bars fault diagnosis for induction machines based on
the wavelet ridge energy spectrum”, Proceedings of the Eighth International Conference
on Electrical Machines and Systems ICEMS 2005, Vol. 3, pp:2274 – 2277, Sept. 2005.
Kohler, J.L., Sottile, J., Trutt, F.C., “Condition Monitoring of Stator Windings in Induction
Motors: Part I-Experimental Investigation of the Effective Negative-Sequence
Impedance Detector”, IEEE Trans. on Ind. Appl., vol. 38, no. 5, pp 1447 – 1453, Sept.
2002.
Lee, S.B., Tallman, R.M., Habetler, T.G.,“A Robust, On-Line Turn-Fault Detection Technique
for Induction Machines Based on Monitoring the Sequence Component Impedance
Matrix”, IEEE Trans. on Power Electron, vol. 18, nº 3, pp. 865-872, May 2003.
Loránd, S., Dobai, J.B., Biró, K.A., “Rotor faults detection in squirrel-cage induction motores
by current signature analysis”, IEEE-TTTC-International Conference on Automation,
Quality and Testing, Robotics, May 13–15, 2004, Cluj-Napoca, Romania.
MathWorks Inc., Wavelet Toolbox User’s Guide, Natick, MA, 2007.
Nandi, S.; Toliyat, H.A., “Condition monitoring and fault diagnosis of electrical machines-a
review”, IEEE Trans. on Energy Conversion, vol. 20, no 4, Dic. 2005, pp 719-729.
Novotny, D.W., Lipo, T.A., Vector Control and Dynamics of AC Drives, Oxford University
Press, New York, 1996.
Supangat R., Ertugrul N., Soong W.L., Gray D.A., Hansen C. and Grieger J., “Detection of
Broken Rotor Bars in Induction Motor using Starting-Current Analysis and Effects
of Loading,” IEE Proceedings - Electric Power Appications, Vol. 153, No. 6, November
2006.
Thomson, W.T., Fenger, M., “Current Signature Analysis to Detect Induction Motor Faults”,
IEEE Industry Applications Magazine, May/June 2001, pp. 26-34.

www.intechopen.com
 Fault Detection
Edited by Wei Zhang

ISBN 978-953-307-037-7
Hard cover, 504 pages
Publisher InTech
Published online 01, March, 2010
Published in print edition March, 2010

In this book, a number of innovative fault diagnosis algorithms in recently years are introduced. These
methods can detect failures of various types of system effectively, and with a relatively high significance.

How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:

Fernando Briz, Michael W. Degner, Pablo Garcia, David Diaz-Reigosa and Alberto Diez (2010). Rotor Fault
Detection in Line-fed Induction Machines Using Complex Wavelet Analysis of Startup Transients, Fault
Detection, Wei Zhang (Ed.), ISBN: 978-953-307-037-7, InTech, Available from:
http://www.intechopen.com/books/fault-detection/rotor-fault-detection-in-line-fed-induction-machines-using-
complex-wavelet-analysis-of-startup-trans

InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
www.intechopen.com