International Conference on Magnetics, Machines & Drives (AICERA-2014 iCMMD)
Condition Monitoring of Induction Motor using
Artificial Neural Network
Ravi C. Bhavsar
Department of Electrical Engineering
UVPCE, Ganpat University
Kherva, India
rcb221087@gmail.com
Rakeshkumar A. Patel
Dr. B.R. Bhalja
Department of Electrical Engineering
UVPCE, Ganpat University
Kherva,India
rap01@ganpatuniversity.ac.in
Department of Electrical Engineering
School of Technology, PDPU,
Gandhinagar,India
bhaveshbhalja@gmail.com
Abstract— This paper deals with stator fault detection of
induction motor. Mathematical modeling of induction motor for
healthy and stator fault condition are explained. In this paper
Artificial Neural Network technique is applied for stator fault
detection in induction motor. By collecting the simulation data
from the mathematical model developed in MATLAB simulink,
ANN is trained. 16 different parameters of induction motor have
been taken to train the neural network. ANN gives best
performance with 10 neurons in hidden layer. The results clearly
show that trained neural network can precisely detect the faults
before any major problem occurs.
Keywords— Condition Monitoring; Induction motor; Analytical
model, Artificial Intelligence, Artificial Neural Network.
I.
INTRODUCTION
Condition monitoring plays a vital role for fault detection
in induction motor. As induction motor is a workhorse for
industry, it is highly preferable to avoid any kind of fault
occurring in the induction motor leading to the breakdown of
the plant. Condition monitoring is a technique or process of
observing the operating characteristics of machine in such a
way that any change in the trend of target parameters could be
used to predict the need for maintenance before any serious
deterioration occurs. Condition monitoring implies the
continuous evaluation of the health of equipments. Previously,
time based maintenance were the mainly used maintenance
strategy [10].
In induction motor the different faults are stator faults,
rotor faults and bearing faults. Stator fault consists 30% , rotor
faults consists 8%, bearing faults consists 40% and other faults
consists 22% of the total faults. Unbalance in voltage, current
and air gap and increment in torque oscillations, losses and
heat are the some of the symptoms of induction motor faults
[5].
The conventional techniques used for fault detection are
magnetic flux, thermal monitoring, torque monitoring, noise
monitoring, partial discharge and vibration monitoring.
Nowadays signal processing techniques like Motor current
Signature Analysis, Fast Fourier Transform, Wavelet Analysis
and Artificial Intelligence techniques are widely used for
condition monitoring of induction motor. Due to its ability to
978-1-4799-5202-1/14/$31.00 ©2014 IEEE
detect the most common machine faults with ease, the MCSA
is regarded as the most popular fault detection technique in
most of the situations [8]. Combination of MCSA and
vibration analysis with Wavelet transform has been
represented in [11, 12] and has been shown to give better
results than Fast Fourier Transform.
Artificial Neural Network, Fuzzy logic and Genetic
algorithm are the techniques of Artificial Intelligence. A
methodology for monitoring and diagnosis of three phase
induction motors external faults has been described and
illustrated in [13]. The paper proposes methodology based on
ANN. A brief description of various AI techniques
highlighting the merits and demerits of each of them has been
given in [6]. The futuristic trends are also discussed.
A neural network based incipient fault detector for small
and medium size induction motors has been developed in [14].
The neural network based incipient fault detector avoids the
problems associated with traditional incipient fault detection
schemes by employing more readily available information
such as rotor speed and stator current. The results of this
evaluation indicate that the neural network based incipient
fault detector provides a satisfactory level of accuracy, greater
than 95%, which is suitable for actual applications.
Condition monitoring of motors through intelligent
diagnostics system based on stator current analysis and PLC
has been developed in [15]. This method is validated by
performing experiments on five different defect levels. In [16],
an overview of condition monitoring technique for induction
motor in precise manner has been given. The authors explained
about the different types of faults occurring in induction motor
and also explained their causes and the percentage as per
authorized literature. The authors have also compared the
different condition monitoring techniques with their
advantages and disadvantages. In [9] a non intrusive approach
for fault detection and diagnosis scheme of bearing faults for
three phase induction motor using stator current signals with
particular interests in identifying the outer race defect at an
early stage has been developed. The Empirical Mode
Decomposition Technique is proposed for analysis of non
stationary stator current signals.
International Conference on Magnetics, Machines & Drives (AICERA-2014 iCMMD)
The most important advantage of the neural network is that
it learns and does not need to be reprogrammed, thereby being
able to perform such tasks where a linear program would fail.
Due to the above advantages ANN is selected for present
paper.
A neural network can perform tasks that a linear program
cannot. The most important advantage of the neural network is
that it learns and does not need to be reprogrammed. Due to
the above advantages ANN is selected for present paper. The
present paper discusses the mathematical modelling of
induction motor and prediction of stator fault by ANN
technique using the simulation of mathematical model.
This paper is organized as follows. Section II describes the
mathematical modelling of induction motor in healthy as well
as in stator fault condition. Section III represents the
simulation of the induction motor with stator fault condition
developed in MATLAB simulink. Simulation results are
discussed in section IV. In section V a method is proposed for
detection of stator fault and neural network is trained for stator
fault detection.
II.
Constant coefficients are given in Appendix I.
• Equation of Torque:
• Equation of Speed:
B. Modelling of induction motor with stator fault
Assume that there is a fault in a stator winding in any one
phase [2].
So, the above equations get modified as follows:
• Equations of Stator:
MATHEMATICAL MODELLING OF INDUCTION MOTOR
To carry out a dynamic modelling of induction motor, we
need to concentrate on the basic equations of induction motor.
As mentioned in various literatures, the equations of healthy
and faulty motors can be represented as follows.
The dynamic model of Induction motor in d-q stationary
reference frame can be described as follows: [1][3][7]
A. Modelling of healthy induction motor
• Equations of Stator:
Where if represents short circuit current and µ denotes the
fraction of short turns and it can be utilized to obtain the
number of inter turns of short circuit winding. It is defined by
Where nf is the number of inter turns short circuit windings
and ns is total number of inter turns in healthy phase. In this
case it should be noted that rotor equations remain unchanged
[4].
As for the equation of short circuit winding flux, we have:
• Equations of Rotor:
• Equations of Current:
• Equations of Current:
if = (−λas2 + (a12iqs + a13iqr) cosθ + (a12ids + a13idr) sinθ) / a11
(19)
International Conference on Magnetics, Machines & Drives (AICERA-2014 iCMMD)
IV.
The constant coefficients of the above equations are given
in Appendix I.
• Equation of Torque:
SIMULATION RESULTS
The motor is simulated for healthy condition at 0% load
(no load), 20% load, 40% load, 80% load and at 100% load
(full load). The motor is also simulated for stator fault with
3turns shorted, 5 turns shorted, 7 turns shorted and 9 turns
shorted.
The parameters of the motor are given in table I.
III.
SIMULATION OF MATHEMATICAL MODEL
A. Simulation of induction motor with stator fault:
The model is composed by five blocks and a MATLAB
code is composed to initialize induction parameters and
process simulated data. Three phase voltage is generated and
then is transferred to axis in abc to dq axis block. The inside
diagrams of two major blocks, dq axis of stator and rotor are
shown in following figures. Equation (1-4) and equation (5-8)
are solved in d axis and q axis blocks. Rotor speed and torque
are calculated in rotor block. Finally, the simulated current is
transferred from dq to abc axis in dq to abc block.
Simulation is carried out at different load condition i.e. 0%
load (no load), 20% load, 40% load, 80% load and 100% load
(full load).
Table I.
Parameters of motor
Output power
33 kW
Frequency
50 Hz
Rated current
46.8 A
Line Voltage
460 V
Number of Poles
4
Resistance of stator
0.087 Ω
Resistance of rotor
0.228Ω
Leakage Inductance of stator
0.0009613 H
Leakage Inductance of rotor
0.0009613 H
Magnetizing Inductance
0.04163 H
Rotor Inertia
0.089 kg.m2
For the proposed technique, the samples are taken for
different parameters like three phase stator current, active
power, reactive power, THD in stator current, THD in voltage,
and speed. There are total 16 parameters. Total 12500 samples
have been taken at five different load conditions for healthy
and faulty motors. The results for healthy condition and faulty
condition are shown as under.Fig. 2 and fig. 3 show the stator
current of motor for healthy and stator fault at full load
condition. From the waveforms it is clearly seen that when
fault occurs the stator current become unbalanced.
Fig. 1. Mathematical model of induction motor with stator fault
Fig. 2. Three phase stator current of healthy motor at full load
International Conference on Magnetics, Machines & Drives (AICERA-2014 iCMMD)
Fig. 3. Three phase stator current of faulty motor at full load with 3 turns
shorted
Fig. 6. Speed of healthy motor at full load
Fig. 4 and fig. 5 show the torque of motor for healthy and
stator fault at full load condition. Under the fault condition the
oscillations are increased in the waveform of the torque.
Fig. 7. Speed of faulty motor at full load with 3 turns shorted
Fig. 4. Torque of healthy motor at full load
When the fault occurs in the induction motor, the THD
increases in stator current. It happens because stator current
becomes unbalanced. From the above simulation of
mathematical models the samples for different parameters are
collected to train the neural network.
V.
ARTIFICIAL NEURAL NETWORK FOR STATOR FAULT
DETECTION
In order to use ANN for identifying induction motor fault
and no fault conditions, it is necessary to select proper inputs
and outputs of the network, structure of the network, and train
it with appropriate data. The objective of training the network
is to adjust the weights so that application of a set of inputs
produces the desired set of outputs. For updating the weights
of the network, back propagation algorithm is used.
Fig. 5. Torque of faulty motor at full load with 3 turns shorted
Fig. 6 and fig. 7 show the speed of motor for healthy and
stator fault at full load condition.
After collecting the samples from the simulation of the
induction motor for healthy and stator fault condition Artificial
Neural Network is trained by those samples. In this case two
layered feed forward neural network is used with one hidden
layer and one output layer. For hidden layer network LogSig
activation function is used while for output layer, purelin
International Conference on Magnetics, Machines & Drives (AICERA-2014 iCMMD)
activation function is used. From the total number of samples,
30% samples i.e. 3750 samples are used to train the neural
network and 70% i.e. 8750 samples are used to test the neural
network. MATLAB function is used for random selection of
training samples and testing samples. The ANN indicates 0 in
the output in healthy condition, and 1 for faulty condition.
Mean squared error is selected for the performance
evaluation of proposed technique. For minimization of mean
squared error, gradient descend rule is applied. Fig.8 shows
the graph of the mean squared error when the number of
neurons are 10 for hidden layer.
Fig.9 shows of plot of the actual output of ANN to predict the
condition of the motor. Output shows that ANN predicts the
condition of motor with least error.
CONCLUSION
In this paper the mathematical modelling of induction
motor in both healthy as well as fault condition is developed.
When fault occurs in the motor the stator current becomes
unbalanced and if the number of shorted turns increases the
severity of the fault is also increased. Number of induction
motor parameters and number of neurons in hidden layer
affects on accuracy of fault detection. Results shows that
accuracy of proposed technique is more than 99%. The output
of the ANN shows that proposed technique predicts the
condition of motor with precision.
In future proposed technique can be used for rotor fault
detection.
APPENDIX I
The following are the equation coefficients used in
mathematical modeling of induction motor.
a 0 = Ls Lr − L2m
a2 =
Lm
a0
a4 =
Ls
a0
a6 = 1 −
Fig.8. Mean squared error plot when number of neuron are 10 for hidden
layer
a8 =
When the neural network was not trained, it was not able to
detect the condition of the induction motor. Once the neural
network is trained, it predicts the condition of motor.
L2m
L s Lr
Lm a 7
Lr
2 μLs
3 a0
a12 = μLs
Lr
a0
1
a3 =
Lls
a1 =
a5 =
1
Llr
a7 =
1
a 6 Ls
a9 =
1
a 6 Lr
2 μLm
3 a0
a13 = μLm
a10 =
a11 =
REFERENCES
[1]
[2]
[3]
[4]
Fig.9. Actual output of ANN
P.C. Krause, O. Wasynczuk, S.D. Sudhoff, “Analysis of Electric
Machinery and Drive Systems”, IEEE Press, New York, 1996.
R.M. Tallam, T.G. Habetler, and R.G. Harley, "Transient Model for
Induction Machines with Stator Winding Turn Faults”, IEEE
Transactions on Industry Applications, Vol. 38, no. 3, pp. 632-637, 2002.
Hamid Fekri Azgomi, Javad Poshtan, “Induction motor stator fault
detection via Fuzzy Logic”, 21st Iranian Conference, IEEE , 2013.
F. Duan, Rastko Zivanovic, “A model for induction motor with stator
faults”, Universities Power Engineering Conference, 22nd Australasian,
IEEE, 2012.
International Conference on Magnetics, Machines & Drives (AICERA-2014 iCMMD)
[5]
S.S. Moosavi, Ait-Amirat, Davod.Arab.Khaburi, “Fault detection in three
phase traction motor using Artificial Neural Networks”, Transportation
electrification conference and expo, IEEE, 2012.
[6] Arfat Siddique, G. S. Yadava, and Bhim Singh, “Applications of
Artificial Intelligence Techniques for Induction Machine Stator Fault
Diagnostics: Review”, 4th IEEE inernational symposium on Diagnostics
for electric machines, power electronics and drives, 2003.
[7] Shakuntla Boora, S.K Agarwal, K.S. Sandhu, “Dynamic d-q axis
modelling of three phase asynchronous machine using MATLAB”,
IJAREEIE,Vol.2, Issue 8, August 2013.
[8] Neelam Mehala, Ratna Dahiya, “Condition monitoring methods, failure
identification and analysis for induction machines”, International Journal
of ciruits, systems and signal processing ,Vol.3, Issue 1, 2009.
[9] Shady S. Refaat , Haitham Abu-Rub , M. S. Saad , E. M. Aboul Zahab
and Atif Iqbal, “ANN- based for detection, diagnosis the bearing fault for
three phase induction motors using current signal”, IEEE International
conference on Industrial Technology, 2013.
[10] Y. Han and Y. H. Song, “Condition Monitoring Techniques for Electrical
Equipment- A Literature Survey”, IEEE Transactions on Power Delivery,
Vol. 18, No.1, January 2003.
[11] Neelam Mehla, Ratna Dahiya, “An approach of Condition Monitoring of
Induction Motor using MCSA”, International Journal of Systems
Applications, Engineering & Development, Volume 1, Issue 1, 2003.
[12] Jeevanand S., Abrahm T. Mathew, “Condition Monitoring of Induction
Motors using Wavelet based analysis of vibration signals”, Second
International Conference on Future Generation Communication and
Networking Symposia, 2008.
[13] ElSayed M.Tag Eldin, Hassan R. Emara, Essam M. Aboul –Zahab,
Shady S. Refaat, "Monitoring and Diagnosis of External Faults in Three
Phase Induction Motors Using ANN”, Power Engineering Society
General Meeting, pp 1-7, IEEE 2007.
[14] Mo-yuen Chow, Peter M. Mangum, and Sui Oi Yee, “A neural network
Approach to Real Time Conition Monitoring of Induction Motors”, IEEE
Transaction on Industrial Electronics,Vol. 38, No.6, Dec. 1991.
[15] Muhammad Irfan, Nordin Saad, Rosdiazli Ibrahim, Vijanth S.
Asirvadam, “An Intelligent Diagnostic System for the Condition
Monitoring of AC motors”, IEEE 8th conference on Industrial Electronics
and Applications, 2013.
[16] Muhammad Irfan, Nordin Saad, Rosdiazli Ibrahim, Vijanth S.
Asirvadam, “Development of an intelligent condition monitoring system
for AC induction motors using PLC”, IEEE Business Engineering and
Idustrial Applications Colloquium, 2013.