Sei sulla pagina 1di 5

2013 Eighth International Conference and Exhibition on Ecological Vehicles and Renewable Energies (EVER)

New Method of Data Treatment Related to Induction


Machine Bearing Fault Using Axial Vibration
Analytic Signal
Abla Bouguerne Abdesselam Lebaroud, Aziz Boukadoum
Laboratory LGEC Constantine, Laboratory LGEC Constantine,
Department of Electrical Engineering, University of Constantine Department of Electrical Engineering, University of Skikda
Alegria Department of Electrical Engineering, University of Tebessa
e-mail:bouguerneabla@yahoo.fr e-mail:lebaroud@yahoo.fr,azizboukadoum@yahoo.fr

Ball bearings is the part most sensitive to mechanical defects


Abstract— This paper deals with a new method called Axial due to aging and wear, for example, cracks or spalling appear
Vibration Analytic Signal applied to Treatment of data for on different parts of the bearing. That is why we are interested
diagnosis the bearing faults of induction motor. This method in the detection and classification of such defects.
consists a extracting vibration vectors forms by time-frequency
representation. The calculation method based on the cloud points
dispersion parameter has two main parts; the first part is the The calculation method has two main parts; the first part is the
Hilbert transform of axial vibration signals. The second part is Hilbert transform of axial vibration signals. The second part is
the normalization compared to module of AVAS. This method the normalization compared to module of AVAS. This method
has been applied on the vibration data of a 5.5-kW induction has been applied on the vibration data of a 5.5-kW induction
motor test bench. motor test bench.
Keywords—time-frequency representation; axial vibration
analytic signal; dispersion parameter ; hilbert transform. II. THE BEARING FAULTS.

at the interface between the rotor and the stator, the ball
I. INTRODUCTION bearing is also have a relatively rapid aging. Typically this
type of fault is diagnosed by the spectrum of measurement
Fault diagnosis is often confronted with the problem of acoustic or vibration.
extracting feature vectors that are relevant per vector size
Assuming that the gap is not constant during of the bearing
reduced and the best possible representation of the fault state.
faults. this information was also present in the current
Several approaches have been discussed in the literature is the
most a view pattern recognition [1], principal component spectrum, since any eccentricity produces anomalies in the
analysis (PCA) and time-frequency representation [2]. distribution of the field in the gap [9]. the analytical expression
of frequencies related to the bearing faults according to its
In diagnosis of rotating machinery, vibration analysis is widely mechanical parameters (number of ball bearings, mechanical
known to be one of the most effective techniques. This stems speed of the rotor, the contact angle of the balls with the
from the fact that oscillation is an inherent characteristic of rings...), and gives the following relationship:
rotating machines and different components of these types of
machinery such as shafts, bearings and gears produce vibration
f bng = f s ± m f v (1)
energy with different characteristics. Any deterioration in the Where
condition of such components can affect their vibratory m = 1, 2 ,3 .......
attributes and manifest itself in the vibration signature. This
allows diagnosis of machine faults by analyzing the vibration f v is a characteristic frequency of vibration.
signature of the system. The expressions for the three types of bearing faults suggests as
follows [10]:
For improved and authentic fault diagnosis using vibration external ring fault : f bng = f s ± m f 0 (2)
analysis techniques it is necessary that the acquired vibration
signals be ‘clean’ enough that small changes in signal attributes interior ring fault : f bng = f s ± f r ± m f i (3)
due to an impending fault in any component can be detected.
To tackle this problem, we have developed a method based on ball fault: f bng = f s ± f frame ± m f b (4)
the cloud points dispersion parameter .

978-1-4673-5271-0/13/$31.00 ©2013 IEEE


III. VIBRATION DATA TREATMENT BY HILBERT
The vibration characteristic frequencies will be calculated TRANSFORM
based on the dimensions of a ball bearing, , shown as Fig. 1. Axial Vibration Analytic Signal (AVAS) method uses the
principle of the analytical signal obtained by Hilbert
transform. The Hilbert transform in the time domain
corresponds to a phase shift of the Fourier transform terms.
Hilbert transform finds a companion function y(t) for a real
function x(t) so that [3] :
The Hilbert transform of a signal y t can be written as: ()
y (t ) ⎯⎯→ ~ TH
y (t ) = ~
y Re (t ) + j~
y Im (t ) (9)
~
where yIm (t ) represents the Hilbert transform of the signal
~
yRe (t ) . The signal , meanwhile, is commonly called analytic
signal.
The amplitude A(t ) of the time signal y (t ) is calculated
using the following relation:
A(t ) = ~
y Re (t ) 2 + ~
y Im (t ) 2 (10)
Applying this to our case for the axial vibration signal Va (t ) ,
we get :
~ ~ ~
Va (t ) ⎯TH
⎯→Va(t ) = Va Re (t ) + jVa Im (t ) (11)
Where
Fig. 1. Dimensions of a ball bearing ~
VaIm (t ) represents The Hilbert transform of the axial
The frequencies of bearing faults include rolling the external ~
vibration signal V a Re (t ) .
ring frequency f 0 , the interior ring frequency f i , the balls ~
V a (t ) is usually called the Axial Vibration Analytic Signal
frequency f b and the frame frequency [11-13]. (AVAS).
N b ⎛ Bd ⎞ The amplitude A(t ) is:
fi = f r ⎜1 + cos β ⎟ (5) ~ ~
2 ⎝ Pd ⎠ Va(t ) = Va Re (t ) 2 + V a Im (t ) 2
(12)
N ⎛ Bd ⎞ The real and imaginary axial vibrations are normalized to the
f O = b f r ⎜1 − cos β ⎟ (6)
module of the AVAS :
2 ⎝ Pd ⎠ ~
~ VaRe (t )
Pd ⎡ ⎛ Bd ⎞ ⎤
2
V ′aRe (t ) = (13)
fb = f r ⎢1 − ⎜ cos β ⎟ ⎥ (7) Va(t )
Bd ⎣⎢ ⎝ Pd ⎠ ⎦⎥ ~
~ Va (t )
fr ⎛ Bd ⎞ V ′aIm (t ) = Im (14)
f frame = ⎜1 − cos β ⎟ (8) Va(t )
2 ⎝ Pd ⎠ The average vibrations is given by:
The five parameters that must be known to calculate the
N ~'
bearing faults frequencies are , the ball diameter Bd , the 1 ~
V a

diameter or average diameter of the frame Pd , the number of mV~ 'a


Re
=
NV~ 'a
∑ V a (k )
k =1
'
Re (15)
balls Nb , the contact angle in radians β , and the rotational
N ~'
frequency f r . 1 ~
V a

Ball bearings is the part most sensitive to mechanical defects


mV~ 'a =
Im
N V~ 'a
∑ V a (k )
k =1
'
Im (16)
due to aging and wear, for example, cracks or spalling appear
on different parts of the bearing. That is why we are interested
in the detection and classification of such defects. The normalized characteristic V ′a Im
~
(V~' a ) , consists of a
Re

cloud of points N V~ 'a . These points represent in a normalized


way the instantaneous evolution of the axial vibrations. The
parameter ξ , that represents the cloud dispersion of these
points is defined as :
Nva
( ~
)( ~
ξ = ∑k =1 Ak − V ' aRe,Im Ak − V ' aRe,Im )T
(17)
~
Where Ak is a point with coordinates V ' a Re (k ) ,
~
V ' a Im (k ) .
The parameter ξ is relatively sensitive to different states of
the machine such as defected bearings, broken bars or
unbalanced feed [4].
IV. EXPERIMENT RESULTS Fig. 3. Acquisition system of signal

An acquisition of current signals was carried out on a test Figures 4, 5,6,7 and 8 represent examples of axial vibration
bench that was made of a 5.5kWinduction motor, Fig. 2. signals for healthy and faulted bearing. It is clear that these
figures do not exhibit any interesting feature. From these
representations, we cannot extract any relevant information.
For this reason, we proceed to the processing of data and use of
additional calculation methods mentioned below.

Fig. 2. Test bench (Motor 5.5 kW connected to the load )


a. healthy machine

The data acquisition set consists of vibration signal recording at a. healthy machine
different levels of load (0%,25%, 50%, 75% and 100%). Two
operating conditions for the machine were considered, namely:
healthy bearing and faulty bearing. The fig. 3. represented the
acquisition system of signals.

All acquisitions are carried out in continuous over a period of


five seconds, with a sampling frequency of 20 kHz. Is 100,000
points for each of the measured signals.

b. bearing fault machine

Fig. 4 . Axial vibration signal for load level 0 %


a. healthy machine a. healthy machine

b. bearing fault machine b. bearing fault machine

Fig. 5. Axial vibration signal for load level 25 % Fig. 7. Axial vibration signal for load level 75 %

a. healthy machine a. healthy machine

b. bearing fault machine b. bearing fault machine

Fig. 6. Axial vibration signal for load level 50 % Fig. 8. Axial vibration signal for load level 100 %
we cannot use the vibration signals directly due to their very [2] Lebaroud, Abdesselam, Clerc, Guy " Classification of Induction
low values. We have proposed a method for calculating a Machine Faults by Optimal Time-Frequency representations ", IEEE
parameter very interesting, the dispersion parameter of the Transactionson Industrial Electronic, Vol.57, n°12, pp.4290-4298,2008.
cloud of points ξ. This parameter is used to calculate the TFR [3] Castelli Marcelo, Juan Pablo Fossatti and José Ignacio Terra “Fault
Diagnosis of Induction Motors Based on FFT” Fourier Transform - nal
and extraction feature vectors. Processing, Dr Salih Salih (Ed.), 2012.
[4] Casimir R, «Diagnostic des défauts des machines asynchrones par
Figure 9 shows a representation of the dispersion of cloud reconnaissance des formes», Thèse de doctorat, Ecole centrale de
Lyon, France, 2003.
points of Axial Vibration Analytics Signal (AVAS) for a [5] Atlas L., Droppo J., and McLaughlin J., “Optimizing time-frequency
healthy machine and bearing faults with different charge level distributions via operator theory,” Proc. SPIE, vol. 3162, pp. 161–171,
(0%.25%%, 50%, 75%,100%). We note in the figure 9 that 1997.
the parameter ξ confirms the separability of classes (healthy [6] Davy M. and Doncarli C., “Optimal kernels of timefrequency
bearing class and faulty bearing class). It is also reported that representations for signal classification” inProc. IEEE-SP Int. Symp.
Time-Freq. Time-Scale Anal., 1998, pp. 581–584.
the parameter ξ is not very sensitive to changes in the level of
[7] Wang M., Rowe G. I., and Mamishev A. V., “Classification of power
charge. quality events using optimal time-frequency representations—Part 2:
application,” IEEE Trans. Power Delivery, vol. 19, pp. 1496–1503,
2004.
[8] Y. del Valle et al., « Particle swarm optimization: Basic concepts,
variants and applications in power systems », IEEE transactions on
evolutionary computation, vol.12, no.2, April 2008.
[9] R. R. Schoen ;T.G.Habetler ; F. Kamran ; R. G. Bartheld; “Motor
Bearing Damage Detection Using Stator Current Monitoring”. IEEE
Transaction on Industry Applications, vol 31, N°6, pp. 1274-1279,
November- December 1995.
[10] M. Blِdt et al. “models for bearing damage detection in induction
motors”, IEEE transactions on industrial electronics, vol. 55, no. 4, April
2008.
[11] B. Yazici, and G. B. Kliman, “An adaptive statistical time–frequency
method for detection of broken bars and bearing faults in motors using
stator current,” IEEE Transactions on Industry Applications 35(2), pp.
442–452. 1999.
[12] H. Ocak, and K. A. Loparo, “Estimation of the running speed and
bearing defect frequencies of an induction motor from vibration data,”
Mechanical Systems and Signal Processing 18(3), pp. 515–533. 2004.
[13] W. Zhou; T. G. Habetler; R. G. Harley; “Stator Current-Based Bearing
Fault Detection Techniques: A General Review”, IEEE International
Fig. 9. Representation of dispersion of the cloud points of Symposium on. Diagnostics for Electric Machines, Power Electronics
axial vibration analytical signals (AVAS). and Drives, pp. 7–10, 6-8 Sept 2007.

V. CONCLUSION

In this paper We have proposed a method for calculating a


parameter very interesting, the dispersion parameter of the
cloud of points. Since we cannot use the vibration signals
directly due to their very low values.. This parameter is used
to calculate the TFR and extraction feature vectors. the
simulation results shown a representation of the dispersion of
cloud points of Axial Vibration Analytics Signal (AVAS) for a
healthy machine and bearing faults with different charge level
that the dispersion parameter confirms the separability of
classes (healthy bearing class and faulty bearing class). It is
also reported that the dispersion parameter is not very
sensitive to changes in the level of charge.

REFERENCES

[1] O. Ondel, E. Boutleux, and G. Clerc, “Feature selection by evolutionary


computing: Application on diagnosis by pattern recognition approach,”
in CAINE, S. Dascalu, Ed., 2005, pp. 219–225.

Potrebbero piacerti anche