Hindawi Publishing Corporation

Shock and Vibration

Volume 2016, Article ID 5836717, 12 pages
http://dx.doi.org/10.1155/2016/5836717

Research Article
Fault Diagnosis System of Induction Motors Based on
Multiscale Entropy and Support Vector Machine with Mutual
Information Algorithm

Shuang Pan,1 Tian Han,1 Andy C. C. Tan,2 and Tian Ran Lin3
1
School of Mechanical Engineering, University of Science and Technology Beijing, 30 Xueyuan Road, Haidian District,
Beijing 100083, China
2
Faculty of Engineering, Universiti Tunku Abdul Rahman, Sungai Long Campus, Kajang, Malaysia
3
School of Mechanical Engineering, Qingdao Technological University, 777 Jialingjian Road, Qingdao 266520, China

Correspondence should be addressed to Tian Ran Lin; [email protected]

Received 10 March 2015; Revised 26 August 2015; Accepted 26 August 2015

Academic Editor: Mohammad Elahinia

Copyright © 2016 Shuang Pan et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An effective fault diagnosis method for induction motors is proposed in this paper to improve the reliability of motors using a com-
bination of entropy feature extraction, mutual information, and support vector machine. Sample entropy and multiscale entropy are
used to extract the desired entropy features from motor vibration signals. Sample entropy is used to estimate the complexity of the
original time series while multiscale entropy is employed to measure the complexity of time series in different scales. The entropy
features are directly extracted from the nonlinear, nonstationary induction motor vibration signals which are then sorted by using
mutual information so that the elements in the feature vector are ranked according to their importance and relevant to the faults. The
first five most important features are selected from the feature vectors and classified using support vector machine. The proposed
method is then employed to analyze the vibration data acquired from a motor fault simulator test rig. The classification results con-
firm that the proposed method can effectively diagnose various motor faults with reasonable good accuracy. It is also shown that the
proposed method can provide an effective and accurate fault diagnosis for various induction motor faults using only vibration data.

1. Introduction Electronics Engineers (IEEE) and by Electric Power Research

Institute (EPRI) through General Electric Corporation [2].
Induction motors are one of the most critical machine The study identified that faulty bearings are the most com-
components in industry today. A breakdown of the motor can mon induction motor fault (representing more than 40% of
lead to the shutdown of a whole production line and results motor faults). This is followed by the stator fault (over 28%),
in expensive downtime. It is thus important to have a reliable
rotor fault (over 8%), and other unspecified faults (more than
condition monitoring (CM) and fault diagnosis system in
12%).
place to continuously monitor the healthy condition of such
machine components. Early fault detection of the motor can Several CM techniques such as vibration analysis, acous-
increase the machine performance and availability, reduce tic analysis [3], motor current signature analysis [4–7],
the consequential damage, prolong the machine useful lifes- electromagnetic field monitoring [8], chemical analysis, and
pan, reduce spare parts inventories, prevent unexpected temperature analysis [9, 10] have been developed in the last
breakdown, and ensure timely maintenance schedules [1]. two decades for the detection and diagnosis of the induction
Induction motor fault diagnosis has attracted considerable motor faults. Vibration technique is one of the most fre-
attention from researchers in recent years. quently employed techniques for motor faults detection due
Table 1 summarizes the most typical faults and their to its ease of use, high accuracy, and reliability. Nevertheless,
occurrence possibilities of induction motors from the sta- vibration signals from induction motors are usually nonlinear
tistical studies conducted by the Institute of Electrical and and nonstationary and have weak signal energies. Traditional
2 Shock and Vibration

Table 1: Typical induction motor faults and the corresponding Induction motor
occurrence possibilities. Power Mechanical
supply load
Bearing Stator Rotor Others
faults faults faults MSE MI
IEEE 42% 28% 8% 22% Vibration data Entropy Feature
acquisition calculation selection
EPRI 40% 38% 10% 12%

Induction motor Support vector

condition machine
frequency domain analysis techniques such as fast Fourier
transform (FFT) are thus not suitable to be directly employed Figure 1: The flowchart of the fault diagnosis algorithm for induc-
for the analysis of motor vibration signals. To overcome this tion motors.
problem, other data analysis techniques such as short time
Fourier transform, wavelet transform, and Hilbert-Huang
transform (HHT) techniques as well as artificial intelligent such as artificial neural networks (ANN) and support vec-
techniques such as neural network have been developed. tor machines (SVM) have been developed in recent years
For instance, Talhaoui et al. [11] used a combined FFT and for mechanical fault classification. SVM has recently been
discrete wavelet transform (DWT) technique to evaluate successfully applied to face detection, object detection and
faults due to broken rotor bars of an induction motor. Roveri recognition, pattern recognition, and information and image
et al. [12, 13] proposed a HHT-based method for damage retrieval due to its high accuracy and good generalization
detection of a bridge structure under moving loads. EMD with a smaller number of samples [24]. It is used in this
technique, the main functionality of HHT, has attracted great study for fault diagnosis of an induction motor due to the
attention in recent years and has been widely employed for consideration that large fault samples are difficult to obtain in
fault diagnosis of rotating machinery. EMD technique still practice. An induction motor fault test rig is also built in the
has several shortcomings though such as mode mixing and laboratory to validate the fault diagnosis technique proposed
the end effect. To alleviate the mode mixing problem, an in this paper.
improved method, ensemble empirical mode decomposition The remainder of the paper is organized as follows.
(EEMD), was developed [14], which has been successfully Section 2 briefly discusses the concept of SampEn, MSE,
employed by Lei et al. [15] for fault diagnosis of rotating MI, and SVM algorithms used in this paper and introduces
machinery. On the other hand, Liang et al. [16] used a combi- the fault diagnosis system proposed in the study. Section 3
nation of power spectrum, cepstrum, bispectrum, and neural describes the experimental setup of the induction motor
network for the data analysis and fault diagnosis of induction fault simulation test rig. Section 4 presents the analysis of
motors. Seera et al. [17] employed a hybrid computer model the experimental data for motor fault diagnosis using the
comprising a Fuzzy Min-Max (FMM) neutral network and a proposed method. The main findings from this study are
Classification and Regression Tree (CART). summarized in Section 5.
Entropy, a measure of the uncertainty of a process [18]
which was first applied in thermodynamics, has been suc-
cessfully employed to analyze the physiological, biological, 2. The Procedure of the Proposed Motor Fault
and electrocardiogram signals [19]. Lin et al. [20] utilized a Diagnosis Method
multiscale entropy method combining with wavelet denois-
ing to detect a motor shaft misalignment. Zheng et al. [21] The flowchart of the proposed induction motor diagnosis
developed a rolling bearing fault diagnosis method based algorithm is shown in Figure 1. The first step of the algorithm
on local characteristic-scale decomposition (LCD) and fuzzy is to calculate the entropy values from the vibration data
entropy (FuzzyEn). Zhu et al. [22] proposed a feature extrac- acquired from an induction motor utilizing SampEn and
tion method based on hierarchical entropy (HE) to effectively MSE. Mutual information technique is then employed to
identify various faulty bearing conditions. Sample entropy select the effective entropy features in the second step. SVM is
(SampEn) and multiscale entropy (MSE) were introduced in applied to discriminate the motor faults in the last step from
their paper to extract features from the vibration data. the selected entropy features.
The variation of parameters used in the entropy algo- For a better understanding of the theoretical foundation
rithms can lead to redundant entropy features causing and process, the procedure and techniques involved in each
excessive training time and affect the system accuracy. To step of the proposed fault diagnosis algorithm are briefly
overcome this problem, a mutual information (MI) technique discussed and explained in the following text.
originally proposed by [23] is adopted in this study to select
the most effective features from the extracted entropies to 2.1. Sample Entropy. Online diagnosis systems are gaining
improve the efficiency and accuracy of the induction motor popularity in industry applications nowadays due to their
fault diagnosis process. capability to detect incipient faults at an early stage. However,
Once a fault feature vector has been selected, a multifault directly measured CM signals are not suitable for online
classifier is needed to identify the faulty conditions of an analysis as a small dataset is not good enough for a reliable
induction motor. Several pattern recognition techniques diagnosis. On the other hand, a larger dataset poses a burden
Shock and Vibration 3

for data processing. Feature extraction is thus a critical initial Xim Xjm
step in a monitoring and fault diagnosis system. The accuracy
of feature extraction has a direct effect on the accuracy
of the final diagnosis outcome. An important criterion is ±r

Sample data
that a feature extraction method should be able to preserve
the critical information for decision-making. Based on this
consideration, SampEn and MSE algorithms are adopted in
this paper to extract the usable effective features from the ±r
vibration data.
Pincus [25] proposed a statistical measure, Approximate
Entropy (ApEn), to quantify the regularity in a short and 0 1 2 i i+1 j j+1 N
noisy time series. ApEn reflects the likelihood of sequences Sample number
that are close to each other. That is to say, within given
Figure 2: A schematic illustration of the distance calculation (𝑚 =
tolerance r, m consecutive data points remain close when one 2).
more data point is known [26]. In general, lower values of
ApEn reflect more regular time series, while higher values are
associated with less predictable time series. ApEn measures
calculation is shown in Figure 2. The distance
the time series for similar segments, and a large number
between the two vectors is calculated using
of such segments yield a small value of ApEn. Importantly,
the ApEn algorithm counts each sequence as matching itself 󵄨 󵄨
𝑑 [𝑋𝑖𝑚 , 𝑋𝑗𝑚 ] = max [󵄨󵄨󵄨𝑥 (𝑖 + 𝑘) − 𝑥 (𝑗 + 𝑘)󵄨󵄨󵄨] ,
to avoid the occurrence of ln(0) in the calculations. This 𝑘=0,1,...,𝑚−1
condition has led to bias in the calculation of ApEn. In (2)
practice, the bias causes ApEn to lack two important expected 𝑖, 𝑗 = 1, 2, . . . , 𝑁 − 𝑚, 𝑖 ≠ 𝑗.
properties: (1) ApEn is heavily dependent on the record
length and is uniformly lower than the expected value for (3) Let 𝑛𝑖𝑚 (𝑟) be the number of distances within r and let
short records and (2) it lacks relative consistency; that is to 𝐵𝑖𝑚 (𝑟) be the total number of m-dimension matched
say, if ApEn of one dataset is higher than that of another, it vector pairs:
should, but does not, remain higher for all tested conditions.
To overcome the shortcomings of ApEn, Richman and 𝑛𝑖𝑚 (𝑟)
𝐵𝑖𝑚 (𝑟) = , 𝑖 = 1, 2, . . . , 𝑁 − 𝑚. (3)
Moorman [27] developed and characterized a new family (𝑁 − 𝑚 − 1)
of statistical measures, sample entropy (SampEn), which
do not count the self-matches. SampEn is precisely the (4) Define the average value of 𝐵𝑖𝑚 (𝑟) such that
negative natural logarithm of the conditional probability
that two sequences with similar m points remain similar 𝑁−𝑚
𝐵𝑖𝑚 (𝑟)
at the next point, where self-matches are not included in 𝐵𝑚 (𝑟) = ∑ . (4)
𝑖=1 (𝑁 − 𝑚)
calculating the probability. Thus, a lower value of SampEn
also indicates more self-similarity in a time series. In addition
to the benefit of eliminating self-matches, SampEn algorithm (5) Repeat steps (1)–(4) for 𝑚 + 1, and 𝐵𝑚+1 (𝑟) can be
is also simpler and uses only about one-half of the time obtained.
compared with ApEn algorithm for the same calculation. (6) For finite N, SampEn is defined as the logarithm of the
SampEn is largely independent of the data record length and ratio of 𝐵𝑚+1 (𝑟) and 𝐵𝑚 (𝑟):
displays relative consistency under various circumstances.
Mathematically, SampEn algorithm can be summarized as 𝐵𝑚+1 (𝑟)
follows. SampEn (𝑚, 𝑟, 𝑁) = − ln [ ]
𝐵𝑚 (𝑟) (5)
Let X be a time series of length 𝑁 : 𝑋 =
{𝑥(1), 𝑥(2), . . . , 𝑥(𝑁)}: 𝑚
= ln 𝐵 (𝑟) − ln 𝐵 𝑚+1
(𝑟) ,

(1) Construct template vectors with dimension m by where 𝑁 is the length of the time series, 𝑚 is the
using the following equation: dimension of sequences to be compared, and r is the
tolerance level for accepting matches. It is convenient
to set the tolerance as 𝑟 × SD, where SD is the
𝑋𝑖𝑚 = [𝑥 (𝑖) , 𝑥 (𝑖 + 1) , . . . , 𝑥 (𝑖 + 𝑚 − 1)] , standard deviation of the dataset. It is worth noting
(1)
𝑖 = 1, 2, . . . , 𝑁 − 𝑚. that the parameters m, r, and N must be fixed for each
calculation.

(2) A match occurs when the distance between two 2.2. Multiscale Entropy (MSE). Regularity in a time series
template vectors (𝑋𝑖𝑚 , 𝑋𝑗𝑚 ) is smaller than predefined is sometimes presented as coarse granularity; thus, the time
tolerance r. A schematic illustration of the distance series needs to be explored using different scales. Costa et al.
4 Shock and Vibration

x1 x2 x3 x4 x5 x6 xi xi+1

Scale 2

··· xi + xi+1
y1 y2 y3 yj =
2

x1 x2 x3 x4 x5 x6 xi xi+1 xi+2

Scale 3
···
y1 y2 xi + xi+1 + xi+2
yj =
3

Figure 3: A schematic illustration of the coarse-graining procedure.

[28] suggested that it is generally difficult to distinguish the Mutual information (MI) is a measure of the variables’
interbeat interval time series of different diseased and healthy mutual dependence. One major advantage of MI for fea-
states when only a single-scale SampEn is used. Therefore, the ture selection is its ability to detect nonlinear relationships
concept of MSE is proposed to resolve this difficulty. For the between variables. MI can also be defined for groups of
sake of brevity, only the two main steps of MSE algorithm are variables (or equivalently for multidimensional variables),
discussed here: which allows one to take the joint relevance and redundancy
of features into account during the feature selection process.
(1) A coarse-graining step, which is used to obtain the Mutual information 𝐼(𝑋, 𝑌) is the amount of uncertainty in
representations of the original time series on different X due to the knowledge of 𝑌. Mathematically, MI can be
time scales. calculated as [23]
(2) The SampEn step, which is used to quantify the
𝑝 (𝑥, 𝑦)
regularities of the coarse-grained time series. 𝐼 (𝑋, 𝑌) = ∑𝑝 (𝑥, 𝑦) log , (7)
𝑥,𝑦 𝑝 (𝑥) 𝑝 (𝑦)
To obtain the coarse-grained time series at a scale factor,
𝜏, the original time series is divided into nonoverlapping where 𝑝(𝑥, 𝑦) is the joint probability distribution function
windows of length 𝜏, and the data points inside each window of 𝑋 and 𝑌 and 𝑝(𝑥) and 𝑝(𝑦) are the marginal probability
are averaged. A coarse-grained time series in MSE is defined distribution functions for 𝑋 and 𝑌. MI is employed in this
using study to select a subset of relevant features, which considers
feature-class mutual information to determine an optimal set
𝑗𝜏 of features by sorting entropy features in descending order.
1 𝑁
𝑦𝑗 (𝜏) = ∑ 𝑥 (𝑖) , 1≤𝑗≤ . (6) It is noted that MI has not been employed previously for
𝜏 𝑖=(𝑗−1)𝜏+1 𝜏 induction motor fault diagnosis. It is used in this study to
select the effective entropy features closely related to a motor
For 𝜏 = 1, the coarse-grained time series is the original fault.
time series. As 𝜏 increases, the length of the resulting coarse-
grained time series decreases. Figure 3 describes 𝜏 coarse- 2.4. Support Vector Machine. SVM algorithm is capable of
grained time series divided from the original time series for handling large feature spaces since the training of SVM
a scale factor of 𝜏. In the conventional MSE algorithm, the is carried out in such a manner that the dimension of
MSE at a scale factor of 𝜏 is defined as the SampEn of the classified vectors does not have any distinct influence on the
first coarse-grained time series. The SampEn for each coarse- performance of SVM as it does on the performance of other
grained time series is calculated and plotted as a function of conventional classifiers. This property is very useful for fault
the scale factor 𝜏 in the final step to complete the procedure classification since the number of features for fault diagnosis
of multiscale entropy analysis. need not be limited. Furthermore, SVM-based classifier also
has better generalization properties than other conventional
2.3. Mutual Information for Feature Selection. The extracted classifiers attributed to the minimum structural misclassifi-
SampEn and MSE features are able to discriminate different cation risk when training the SVM classifier, whereas other
types of motor faults. Nevertheless, a feature vector with high traditional classifiers are usually trained to minimize the
dimension will be very time consuming to analysis. There- empirical risk [29, 30]. SVM can also be used in nonlinear
fore, it is our aim to reduce the feature space by selecting only classification tasks with the selection of the kernel functions.
the most effective features containing the fault information The definition of a legitimate kernel function is given by
to improve the performance of SVM classification and the Mercer’s theorem which states that the function must be
effectiveness of fault diagnosis. continuous and positive definite. Any function that satisfies
Shock and Vibration 5

Table 2: Formulation of kernel functions. Control board Belt Fan

Kernel 𝐾 (x𝑖 , x𝑗 )
Linear function 𝐾 (x𝑖 , x𝑗 ) = x𝑖𝑇 × x𝑗
Polynomial function 𝐾(x𝑖 , x𝑗 ) = (𝑔x𝑖𝑇 × x𝑗 + 𝑟)𝑑 , 𝑔 > 0
󵄩 󵄩
− 󵄩󵄩󵄩x𝑖 − x𝑗 󵄩󵄩󵄩
Radial basis function 𝐾 (x𝑖 , x𝑗 ) = exp ( 󵄩 2 󵄩 )
2𝑔

Mercer’s theorem can be used as a kernel function to compute

the dot product in the feature space. Various kernel functions
such as linear, polynomial, and radial basis functions are used
in SVM classification. The selection of an appropriate kernel
function is very important in the classification process as
the kernel defines the feature space in which the training set
examples are classified. Linear, polynomial, and radial basis
functions are evaluated and formulated in this paper and Accelerometer Motor Pulley
listed in Table 2, where parameter 𝑔 is the kernel parameter.
The fault diagnosis of an induction motor is a multifault Figure 4: A description of the fault simulation test rig and the
classification since there are various fault types for the motor instrumentation.
(also see Table 1). Commonly used techniques for multiclass
classification include “one-against-all” or “one-against-one”
techniques. The “one-against-one” technique is adopted in
this paper. For this technique, if 𝑘 is the number of classes, with changeable pitch blades, bearings, and the supporting
it constructs 𝑘(𝑘 − 1)/2 classifiers and each one is trained on structure.
the data from two classes. For the training data from the 𝑖th Six 0.5 kW, 60 Hz, 4-pole induction motors (1 healthy
and 𝑗th classes, the two-class classification problem is solved and 5 faulty) were used in the experiment to generate the
as follows: required dataset under the full-load condition. Data from the
1 󵄩󵄩 𝑖𝑗 󵄩󵄩2 𝑇
healthy motor is used as a benchmark for comparison with
Minimize: 󵄩󵄩𝜔 󵄩󵄩 + 𝑐∑𝜉𝑡𝑖𝑗 (𝜔𝑖𝑗 ) , the experimental data from other faulty motors. The faulty
2 󵄩 󵄩
𝑡 motors each have a specific fault: (1) unbalance rotor (MUN),
𝑇 𝑖𝑗 (2) broken rotor bar (BRB), (3) faulty bearing (outer race,
subject to: (𝜔𝑖𝑗 ) 𝜑 (𝑥𝑡 ) + 𝑏𝑖𝑗 ≥ 1 − 𝜉𝑡 , FBO), (4) bowed rotor (BR), and (5) eccentricity (angular
𝑇
misalignments (AMIS) and parallel misalignments (PMIS))
𝑖𝑗
(𝜔𝑖𝑗 ) 𝜑 (𝑥𝑡 ) + 𝑏𝑖𝑗 ≥ 𝜉𝑡 − 1, as shown in Figure 5. Furthermore, phase unbalance (PUN)
is also simulated in the experiment by adding windings on
𝑖𝑗
𝜉𝑡 ≥ 0, one of the phases of the healthy motor (NOR). The faulty
(8) conditions of these motors are also listed in Table 3.
Three accelerometers are used in the experiment to mea-
where parameter 𝑐 is the penalty parameter of the error sure the vibration signals in the three coordinate directions
classification. The Max Win Strategy [22] is utilized during (i.e., horizontal, vertical, and axial) of the motors to evaluate
the effectiveness and the functionality of the proposed fault
the classification as follows: if sign((𝜔𝑖𝑗 )𝑇 𝑓(𝑥𝑡 ) + 𝑏𝑖𝑗 ) indicates
diagnosis algorithm. The frequency range of the vibration
that 𝑥 is in the 𝑖th class, the vote for the 𝑖th class is increased
signals in the setup is up to 5 kHz and the total number
by one; otherwise, the 𝑗th class is increased by one. Then, 𝑥
of data samples acquired is 16384. Figure 6 displays a set of
is designated to be in a class with the maximum number of
time waveforms of the measured vibration signals for the
votes.
simulated induction motor faults investigated in this study.
The fault diagnosis procedures described in this section
It is observed that the difference between the vibration time
will be employed to analyze the experimental data from an
waveforms for the normal, unbalance rotor, broken rotor bar,
induction motor fault simulation test rig to be elaborated in
and phase unbalance cases is insignificant. For these cases,
the following sections.
the waveforms appear to be slightly distorted from a pure
sine wave having the fundamental frequency of the shaft
3. A Description of the Fault Simulation Test rotating speed. For the eccentricity cases, the waveforms are
Rig and the Experimental Setup mainly composed of the vibration from the fundamental shaft
rotating frequency and are superimposed with vibration from
The induction motor test rig was designed to carry out higher order components. For the bowed rotor and faulty
the fault simulation as shown in Figure 4. It consists of an bearing cases, the vibration waveforms are largely distorted
induction motor, a pulley and pulley belt, a shaft and a fan by impact components.
6 Shock and Vibration

Rotor unbalance Broken rotor bar

Faulty bearing Bowed rotor Eccentricity

Figure 5: Typical induction motor faults.

Normal

Rotor unbalance

Broken rotor bar

Phase unbalance

Angular misalignments

Parallel misalignments

Bowed rotor

Faulty bearing

0 0.25 0.50
Time (s)

Figure 6: The representative vibration waveforms of the induction motor faults.

Shock and Vibration 7

Table 3: A description of the faulty induction motors.

Fault condition Fault description Others

Unbalance rotor (MUN) Unbalance mass on the rotor: 8.4 g
Broken rotor bar (BRB) Number of broken bars: 12 Total number of 34 bars
Faulty bearing (FBO) Spall on the outer race NSK-6203
Bowed rotor (BR) Rotor deflection: 0.075 mm Air gap: 0.25 mm
Eccentricity (AMIS) Angular misalignments Adjusting the bearing pedestal
Eccentricity (PMIS) Parallel misalignments Adjusting the bearing pedestal
Phase unbalance (PUN) Adding windings on one of the phases Using the normal motor

Consistency analysis Consistency analysis

2.5 2.5

FBO
2 2
BR
SampEn

FBO
AMIS PMIS
AMIS
MUN NOR PUN MSE
PMIS MUN NOR
1.5 1.5 BR
BRB PUN
BRB

1 1
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Sample number Sample number

Upper limit Upper limit

Mean Mean
Lower limit Lower limit
(a) (b)

Figure 7: Consistency analysis of the two chosen entropy features. (a) Consistency analysis of SampEn (tolerance, 𝑟 = 0.2) and (b) consistency
analysis of MSE (scale factor, 𝜏 = 2).

4. Application of the Proposed Algorithm for fault, and the green line represents the mean entropy value for
Motor Fault Diagnosis each fault. The analysis of the distribution of the two entropy
features using the “3s” criteria indicates that both SampEn
4.1. Consistency Analysis for Each Fault Condition. An effec- and MSE satisfy the consistency requirement.
tive feature should be able to distinguish different induction
motor fault conditions and satisfy the consistency require- 4.2. Entropy Calculation and Feature Selection. As described
ment for each induction motor fault. A consistency analysis is in Section 2, three parameters, namely, the tolerance r, the
thus required before the use of entropy features for induction dimension of sequences m, and the length of the time series N,
motor fault diagnosis since motor vibration data typically are used in the calculation of SampEn. The length of each data
contains nonlinear components and noise. Random errors sample is automatically determined by the data acquisition
in the experiment can cause variation of the calculated setup, which is 𝑁 = 4096. The dimension of sequences,
entropy feature values. A large variation of the calculated 𝑚 = 2, is used in this calculation. The tolerance r is commonly
entropy values will render the features inappropriate for fault expressed as a fraction of the standard deviation (SD) of
classification. The “3s” rule in statistics is adopted in this the time series that affects the SampEn value. A small 𝑟 will
study to evaluate the consistency of the two entropy features, lead to a poor conditional probability, while a large r will
SampEn and MSE. result in losing too much system information. In principle,
Figure 7 shows the consistency analysis of the two entropy r should be a value that minimizes the effect of noise on
features for the simulated motor faults. The black dots are the the calculation of the SampEn. The value of r chosen in this
entropy values of 20 data samples for each simulated motor study ranges between 0.1SD and 0.7SD. Figure 8 illustrates
8 Shock and Vibration

3 Table 4: Entropy features selection using MI algorithm.

Features number Entropy Dimension Tolerance Scale factor

2.5
1 MSE 𝑚=2 𝑟 = 0.2 𝜏=1
2 MSE 𝑚=2 𝑟 = 0.2 𝜏=2
2
3 MSE 𝑚=2 𝑟 = 0.2 𝜏=3
SampEn

4 MSE 𝑚=2 𝑟 = 0.2 𝜏=5

1.5
5 MSE 𝑚=2 𝑟 = 0.2 𝜏=7
1

0.5 Table 5: Classification accuracy using different SVM kernel func-

tions.
0 Parameters of Classification
0.1 0.2 0.3 0.4 0.5 0.6 0.7 Number Kernel function
svmtrain accuracy
Tolerance
1 Linear function -𝑐8-𝑔4-𝜏0 91.25% (73/80)
AMIS MUN Polynomial
BR NOR 2 -𝑐8-𝑔4-𝜏1 91.25% (73/80)
function
BRB PMIS
PUN 3 Radial basis
FBO -𝑐8-𝑔4-𝜏2 96.25% (77/80)
function
Figure 8: The effect of chosen tolerance values on the entropy
feature, SampEn.
case is the lowest under this condition. It is further shown that
the MSE values for the BR and FBO cases decrease rapidly as
the effect of r value on the calculated SampEn value for the the scale factor increases due to the reduced complexity of the
motor faults simulated in this study. It is shown that the waveforms (see Figure 9). Conversely, the MSE values for the
value of SampEn decreases steadily for increasing 𝑟 value. The AMIS and PMIS cases increase as the scale factor increases
calculated SampEn for the faulty bearing (FBO) case serves attributed to the increased complexity of the waveforms (see
as the upper bound for all cases attributed to the impact Figure 9). Starting from the scale factor, 𝜏 = 4, onward, the
impulses and high frequency components in the vibration MSE values of the coarse-grained time series for AMIS and
signal (see Figure 6). On the contrary, the calculated SampEn PMIS cases are much larger than those of PUN and MUN
for the broken rotor bar (BRB) case serves as the lower cases for the same scale factor. The features for these faulty
bound of all cases as a result of the least noisy low frequency cases can be clearly distinguishable for higher scale factors.
vibration waveform recorded for this case (Figure 6). It is This is particularly so for the three former cases. Therefore,
further observed that the calculated SampEn for the bowed these MSE features will be adopted to distinguish the fault
rotor (BR), BRB, FBO, and normal (NOR) cases can be conditions among AMIS, PMIS, and PUN cases in the fault
clearly distinguished from each other and thus can be used classification.
directly for motor fault classification. On the other hand, the In the next step, MI algorithm is applied to sort the order
entropy values for the eccentricity (AMIS, PMIS), unbalance of the features according to their importance and relationship
rotor (MUN), and phase unbalance (PUN) cases are clustered with the faults. The first five features are selected and listed in
together which are difficult to distinguish from each other. Table 4.
Further efforts are thus required to derive a more suitable
(separable) feature for these later cases for fault classification. 4.3. Fault Classification. Three different kernel functions
MSE features are therefore selected to serve this purpose including linear function, polynomial function, and radial
to overcome the limitation of the SampEn features. The basis function of SVM are used in the fault classification
parameter values used in MSE feature extraction are 𝑚 = 2, in this study. A supervised mode is used for data training
𝑟 = 0.2SD, and 𝑁 = 4096 with a maximum scale factor using the LIBSVM Matlab Toolbox. Figure 11 illustrates the
of 𝜏 = 10. Figure 9 compares the time waveforms of the classification accuracy using different kernel functions and as
coarse-grained time series for AMIS and FBO cases using a function of the selected parameter value 𝑔 in the training.
the first five scale factors: 𝜏 = 1, 2, . . . 5. It is shown that The classification efficiency for the three SVM models using
the AMIS waveform becomes more complex with increasing the parameter values, 𝑐 = 8 (a cost parameter) and 𝑔 = 4
scale factor while the FBO waveform is less complex for larger (a width parameter), is given in Table 5 for comparison. It is
scale factors. shown that the SVM model using the radial basis function
Figure 10 compares the MSE values for the simulated predicts the measured values with the highest accuracy.
motor faults investigated in this study for the scale factor After the data training, the second phase of SVM algo-
ranging from 1 to 10. For the scale factor of 1, the calculated rithm is to test and validate whether the classifier has learnt
MSE value is the same as the SampEn value. The MSE value from unseen examples in the training phase. A second-fold
for the FBO case is the highest and the MSE value for the BRB cross-validation method is applied to the 160 experimental
Shock and Vibration 9

𝜏=1

𝜏=2

𝜏=3

𝜏=4

𝜏=5

AMIS FBO

Figure 9: Time waveforms of coarse-grained time series.

Table 6: The second-fold test accuracy of the proposed method.

Dataset (160)
Correctly classified Wrongly classified Classification accuracy
Training set Test set
Set-1 (80) Set-2 (80) 74 6 92.50%
Set-2 (80) Set-1 (80) 76 4 95.00%
Average classification accuracy (all ten features) 93.75%
Set-1 (80) Set-2 (80) 77 3 96.25%
Set-2 (80) Set-1 (80) 77 3 96.25%
Average classification accuracy (the first five features using MI) 96.25%

datasets for the validation of the proposed induction motor method. Only three wrongly classified cases are found from
fault diagnosis system. In a 𝑘-fold cross-validation method, the 160 datasets for the MUN and NOR cases due to the
the dataset is divided into 𝑘 subsets, and the holdout method similarity of the waveform (see Figure 6) and entropy values
[31] is repeated 𝑘 times. At each time, 𝑘−1 subsets are used for (see Figure 10) between these two cases.
training and 𝑘th subset is used for testing. The average error Figure 13 compares the classification accuracy of the
for all 𝑘 trials is computed. The second-fold test accuracy proposed method and another method previously developed
of the proposed method is shown in Table 6. The average by the same authors [1]. It is shown that the method proposed
classification accuracy using the first five features and the MI in this study is as accurate or better for motor fault diagnosis
algorithm is 96.25%. in most cases except for one (NOR). The proposed method
The prediction accuracy of the SVM model using the can identify 8 different operational conditions of induction
radial basis function is graphically illustrated in Figure 12. It is motors with high accuracy which is an advantage over
shown that the SVM classifier can identify the specific motor a previous work [20] where a single fault such as shaft
fault with 100% accuracy for most faults using the proposed misalignment was studied. The proposed method also differs
10 Shock and Vibration

2.4 1

2.2
2 0.95

1.8
1.6 0.9

Accuracy
MSE

1.4
1.2 0.85

1
0.8 0.8

0.6
0.4 0.75
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9
c
Scale factor

AMIS Linear function

MUN
BR Polynomial function
NOR
BRB PMIS Radial basis function
FBO PUN (a) Parameter 𝑐

Figure 10: MSE of vibration signals. 1

0.98

from other previous works for multifault classification [32–

34] by using only vibration data and the two extracted entropy 0.96
Accuracy

features, SampEn and MSE.

0.94
5. Conclusion
A fault diagnosis algorithm using two entropy features 0.92
(SampEn and MSE), MI algorithm, and SVM classifier was
proposed in this paper for induction motor fault diagnosis.
The major steps and their sequential use in this algorithm 0.9
1 2 3 4 5
are (1) extracting the entropy features from the nonlinear, g
nonstationary induction motor vibration signals; (2) sorting
the features by using MI algorithm so that the elements in the Linear function
feature vector are ranked according to their importance and Polynomial function
relevant to the faults; (3) selecting the five most important fea- Radial basis function
tures from the feature vectors and then classifying them using (b) Parameter 𝑔
SVM method. The proposed classification and fault diagnosis
technique was validated using the induction motor vibration Figure 11: Classification accuracy of the three SVM models using
data in the study. The classification results confirmed that different parameter values in the data training.
the proposed fault diagnosis algorithm can effectively detect
8 various faults of the induction motor with an average
accuracy of 96.25%. The algorithm proposed in this study
thus provides an effective tool for a reliable fault diagnosis of
a multifault mechanical system such as an induction motor as proposed in the paper can largely improve the effectiveness
using only vibration CM data. The study presented in this and accuracy of machine fault classification and diagnosis
paper highlights the notions that (1) entropy features are and thus is very useful in the development of automatic fault
effective parameters to extract useful information from the diagnosis systems.
vibration data, which can satisfy the consistency requirement
for induction motor fault diagnosis; (2) mutual information
technique is an effective tool to reduce the feature space Conflict of Interests
extracted from the CM data, which can achieve a higher
recognition rate with selected features; (3) the combination The authors declare that there is no conflict of interests
of entropy, mutual information, and support vector machine regarding the publication of this paper.
Shock and Vibration 11

PUN [4] W. T. Thomson and M. Fenger, “Current signature analysis

PMIS to detect induction motor faults,” IEEE Industry Applications
NOR
Magazine, vol. 7, no. 4, pp. 26–34, 2001.
Output label

[5] W. T. Thomson, D. Rankin, and D. G. Dorrell, “On-line current

MUN
monitoring to diagnose airgap eccentricity in large three-
FBO phase induction motors-industrial case histories verify the
BRB predictions,” IEEE Transactions on Energy Conversion, vol. 14,
BR
no. 4, pp. 1372–1378, 1999.
[6] S. Williamson and K. Mirzoian, “Analysis of cage induction
AMIS
0 10 20 30 40 50 60 70 80 motors with stator winding faults,” IEEE Transactions on Power
Test sample number Apparatus and Systems, vol. 104, no. 7, pp. 1838–1842, 1985.
[7] R. R. Schoen, T. G. Habetler, F. Kamran, and R. G. Bartheld,
Desired “Motor bearing damage detection using stator current moni-
Predicted toring,” IEEE Transactions on Industry Applications, vol. 31, no.
Figure 12: Accuracy of the proposed model for induction motor 6, pp. 1274–1279, 1995.
fault diagnosis. [8] F. Thollon, G. Grellet, and A. Jammal, “Asynchronous motor
cage fault detection through electromagnetic torque measure-
ment,” European Transactions on Electrical Power Engineering,
1 vol. 3, no. 5, pp. 375–378, 1993.
[9] E. R. Filho, R. R. Riehl, and E. Avolio, “Automatic three-phase
squirrel-cage induction motor test assembly for motor thermal
behavior studies,” in Proceedings of the IEEE International
0.75
Symposium on Industrial Electronics (ISIE ’94), pp. 204–209,
Classification accuracy

Santiago, Chile, May 1994.

[10] R. Beguenane and M. E. H. Benbouzid, “Induction motors
0.5 thermal monitoring by means of rotor resistance identification,”
IEEE Transactions on Energy Conversion, vol. 14, no. 3, pp. 566–
570, 1999.
[11] H. Talhaoui, A. Menacer, A. Kessal, and R. Kechida, “Fast
0.25 Fourier and discrete wavelet transforms applied to sensorless
vector control induction motor for rotor bar faults diagnosis,”
ISA Transactions, vol. 53, no. 5, pp. 1639–1649, 2014.
0 [12] N. Roveri and A. Carcaterra, “Damage detection in structures
AMIS BR BRB FBO MUN NOR PMIS PUN under traveling loads by Hilbert-Huang transform,” Mechanical
Systems and Signal Processing, vol. 28, pp. 128–144, 2012.
The proposed method [13] N. Roveri, A. Carcaterra, and A. Akay, “Frequency intermit-
Reference [1] tency and energy pumping by linear attachments,” Journal of
Sound and Vibration, vol. 333, no. 18, pp. 4281–4294, 2014.
Figure 13: Classification accuracy comparison.
[14] Z. Wu and N. E. Huang, “Ensemble empirical mode decom-
position: a noise-assisted data analysis method,” Advances in
Adaptive Data Analysis, vol. 1, no. 1, pp. 1–41, 2009.
Acknowledgment [15] Y. Lei, Z. He, and Y. Zi, “Application of the EEMD method to
rotor fault diagnosis of rotating machinery,” Mechanical Systems
This paper is supported by the Fundamental Research Funds and Signal Processing, vol. 23, no. 4, pp. 1327–1338, 2009.
for the Central Universities (Grant no. FRF-TP-14-018A2). [16] B. Liang, S. D. Iwnicki, and Y. Zhao, “Application of power
spectrum, cepstrum, higher order spectrum and neural net-
References work analyses for induction motor fault diagnosis,” Mechanical
Systems and Signal Processing, vol. 39, no. 1-2, pp. 342–360, 2013.
[1] T. Han, B.-S. Yang, W.-H. Choi, and J.-S. Kim, “Fault diagnosis [17] M. Seera, C. P. Lim, D. Ishak, and H. Singh, “Offline and online
system of induction motors based on neural network and fault detection and diagnosis of induction motors using a hybrid
genetic algorithm using stator current signals,” International soft computing model,” Applied Soft Computing, vol. 13, no. 12,
Journal of Rotating Machinery, vol. 2006, Article ID 61690, 13 pp. 4493–4507, 2013.
pages, 2006. [18] B. Guo, S. Song, A. Ghalambor, and T. R. Lin, Offshore Pipelines,
[2] T. Han, B.-S. Yang, and Z.-J. Yin, “Feature-based fault diagnosis Gulf Professional Publishing, Oxford, UK, 2nd edition, 2014.
system of induction motors using vibration signal,” Journal of [19] M. Costa, A. L. Goldberger, and C.-K. Peng, “Multiscale entropy
Quality in Maintenance Engineering, vol. 13, no. 2, pp. 163–175, analysis of complex physiologic time series,” Physical Review
2007. Letters, vol. 89, Article ID 068102, pp. 1–4, 2002.
[3] Y.-S. Lee, J. K. Nelson, H. A. Scarton, D. Teng, and S. A. [20] J.-L. Lin, J. Y.-C. Liu, C.-W. Li, L.-F. Tsai, and H.-Y. Chung,
Ghannad, “Acoustic diagnostic technique for use with electric “Motor shaft misalignment detection using multiscale entropy
machine insulation,” IEEE Transactions on Dielectrics and Elec- with wavelet denoising,” Expert Systems with Applications, vol.
trical Insulation, vol. 1, no. 6, pp. 1186–1193, 1994. 37, no. 10, pp. 7200–7204, 2010.
12 Shock and Vibration

[21] J. Zheng, J. Cheng, and Y. Yang, “A rolling bearing fault diagno-

sis approach based on LCD and fuzzy entropy,” Mechanism and
Machine Theory, vol. 70, pp. 441–453, 2013.
[22] K. Zhu, X. Song, and D. Xue, “A roller bearing fault diagnosis
method based on hierarchical entropy and support vector
machine with particle swarm optimization algorithm,” Mea-
suremen, vol. 47, no. 1, pp. 669–675, 2014.
[23] G. Doquire and M. Verleysen, “Mutual information-based
feature selection for multilabel classification,” Neurocomputing,
vol. 122, pp. 148–155, 2013.
[24] A. Widodo and B.-S. Yang, “Support vector machine in machine
condition monitoring and fault diagnosis,” Mechanical Systems
and Signal Processing, vol. 21, no. 6, pp. 2560–2574, 2007.
[25] S. M. Pincus, “Approximate entropy as a measure of system
complexity,” Proceedings of the National Academy of Sciences of
the United States of America, vol. 88, no. 6, pp. 2297–2301, 1991.
[26] M. Costa, C.-K. Peng, A. L. Goldberger, and J. M. Hausdorff,
“Multiscale entropy analysis of human gait dynamics,” Physica
A: Statistical Mechanics and its Applications, vol. 330, no. 1-2, pp.
53–60, 2003.
[27] J. S. Richman and J. R. Moorman, “Physiological time-series
analysis using approximate entropy and sample entropy,” Amer-
ican Journal of Physiology—Heart and Circulatory Physiology,
vol. 278, no. 6, pp. H2039–H2049, 2000.
[28] M. Costa, A. L. Goldberger, and C.-K. Peng, “Multiscale entropy
analysis of biological signals,” Physical Review E, vol. 71, no. 2,
Article ID 021906, 18 pages, 2005.
[29] D. Hanbay, “An expert system based on least square support
vector machines for diagnosis of the valvular heart disease,”
Expert Systems with Applications, vol. 36, no. 3, pp. 4232–4238,
2009.
[30] M. Saimurugan, K. I. Ramachandran, V. Sugumaran, and N.
R. Sakthivel, “Multi component fault diagnosis of rotational
mechanical system based on decision tree and support vector
machine,” Expert Systems with Applications, vol. 38, no. 4, pp.
3819–3826, 2011.
[31] R. Kohavi, “A study of cross-validation and bootstrap for
accuracy estimation and model selection,” in Proceedings of the
International Joint Conference on Artificial Intelligence (IJCAI
’95), vol. 2, pp. 1137–1145, Quebec, Canada, August 1995.
[32] G. Niu, T. Han, B.-S. Yang, and A. C. C. Tan, “Multi-agent
decision fusion for motor fault diagnosis,” Mechanical Systems
and Signal Processing, vol. 21, no. 3, pp. 1285–1299, 2007.
[33] G. Niu, A. Widodo, J.-D. Son, B.-S. Yang, D.-H. Hwang, and D.-
S. Kang, “Decision-level fusion based on wavelet decomposition
for induction motor fault diagnosis using transient current
signal,” Expert Systems with Applications, vol. 35, no. 3, pp. 918–
928, 2008.
[34] A. Widodo, B.-S. Yang, and T. Han, “Combination of inde-
pendent component analysis and support vector machines for
intelligent faults diagnosis of induction motors,” Expert Systems
with Applications, vol. 32, no. 2, pp. 299–312, 2007.
 International Journal of

Rotating
Machinery

International Journal of
The Scientific
Engineering Distributed
Journal of
Journal of

Hindawi Publishing Corporation

World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
Sensors
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

http://www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Publishing Corporation Volume 2014
Navigation and
Observation
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
in Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com
http://www.hindawi.com Volume 2014

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014