You are on page 1of 6

2018 International Conference on Computing, Power and Communication Technologies (GUCON)

Galgotias University, Greater Noida, UP, India. Sep 28-29, 2018

Bearing Fault Diagnosis of Induction Motor Using


Thermal Imaging
Anurag Choudhary Shimi S.L Aparna Akula
ME Student Assistant Professor Scientist
Electrical Department Electrical Department Computational Instrumentation
NITTTR, Chandigarh NITTTR, Chandigarh CSIR-CSIO, Chandigarh
anurag.choudhary087@gmail.com shimi.reji@gmail.com aparna.akula@csio.res.in

Abstract—Temperature is one of the most important of applications such as electrical and mechanical maintenance
indicators of health of rotating machines. To provide automatic [5], agriculture [6], defense [7], aviation, geological survey
fault diagnosis using temperature information it is meaningful [8], automotive [9], medical [10], electronics [11] etc. for
to extract optimal featureswhich can be used to identify the fault detection. It is also used in industries to detect serious
potential fault. An automatic bearing fault diagnosis approach faults in equipment, quality control and process control. In
using non-invasive contactless thermal infrared imaging is order to perform automatic health monitoring as well as fault
proposed for early fault diagnosis of an induction motor. The diagnosis of the rotating machines using thermal imaging
authors analyse thermal images of four different bearing three important stages are involved, image acquisition, image
conditions in a three phase induction motor: outer race pre-processing, feature extraction and selection and
defected bearing, inner race defected bearing, lack of
classification.
lubrication and healthy bearing. As a first step thermal images
of induction motor were acquired then preprocessed using 2D- It is very important to understand the factors that
DWT to decomposed the thermal images.This is followed contribute to the variations in the thermal behavior of the
extracting relevant features and selecting the strongest feature bearings. In this work we study these factors by examing the
using Mahanabolis distance critera. Finally the selected thermal images of the bearing region of the induction motor.
features are given to a SVM classifier for classification of The mutual effects of the thermal behaviour and bearing
bearing condition. The experimental results indicate decent health conditions are analyzed accordingly.The
classification performance of the proposed technique for
characteristics of the temperature are stable to wear and tear,
bearing fault detection.
type of fault, and its severity [12]. Signal processing
Index Terms— Thermal Imaging; Condition Monitoring; techniques employed for extracting the features of the
Thermal Camera; Preventive Maintanence. temperature play an important role in the systems intended
for the automated analysis and interpretation of bearing
I. INTRODUCTION faults. Hence, it is necessary to identify suitable signal
processing techniques, which can detect and categorise the
Bearing is an important element used in rotating
fault accurately. In this study, we focus of designing one such
machines to establish a free rotational motion by reducing
automatic bearing fault diagnosis frameworkcapable of
friction on the rotor shaft. Bearings are designed for rotaing
identifying four different bearing states of a induction motor
machines to have a useful and long life. In all practical
(as shown in Fig.1)using thermal images.
scenarios, bearing must be perfectly installed, lubricated and
maintained.Improper handling, poor operating environments This paper is organised as follows: related work using
and contaminated areas invite immature bearing failure. Itis thermal imaging techniques for diagnosis of rotating
important to find the exact cause of bearing failurefor repair machinery faults has been discussed in section II. Then, the
and maintainence. Bearings cause for bearing defects are methodology of bearing fault diagnosis and classification is
loose or tight fits, misalignment, excessive and reverse presented in section III. In the section IV, the experimental
loading, wear and tear, contamination in lubricants, lubricant study of the bearing fault diagnosis has been discussed. The
failure, corrosion, and overheating. Many researchers have experimental results arepresented in section V. Lastly done
investigated that the primary causation of failure of electrical with the conclusion of the proposed work in section VI.
rotating machinery is bearing deficiency [1–2].
Unpredictable damage of bearings may affect other
rotating parts of the machines and also increase the severity
of the plant breakdown [3]. Traditional methods for
inspecting the bearing systems are motor current analysis,
wear debris analysis, vibration monitoring, etc. [4]. Thermal
Imaging Technique (TIT), is emerging as an alternative
technique as it is non-intrusive, contactless, and less complex.
The thermal cameras sense radiation in a long-infrared range
(9-14μm) of the electromagnetic-spectrum, and produces
thermal images, known as thermograms. According to the
Stefan-Boltzmann’s law, temperature measurement depends
on the intensity of the emitted radiation. TIT has a vast range Fig. 1. Fig.1 Bearing in different faulty and healthy condition (a) Healthy
(b) inner race defected bearing (c) outer race defect bearing (d) Lack
of lubrication

978-1-5386-4491-1/18/$31.00 ©2018 IEEE 950


II. THERMAL IMAGING AS A DIAGNOSTICS TOOL Convolution and down sampling process is denoted by *
Diagnosis is the art of health monitoring of machine by and ՝, respectively. Here ൫‫ܮ‬௫ ǡ ‫ܮ‬௬ ൯and ൫‫ܪ‬௫ ǡ ‫ܪ‬௬ ൯ are low-pass
observing its symptoms to identify the cause of failure. It is and high-pass filters.
mainly utilized for monitoring, detection of abnormal The filtering and down sampling process for ‫ݓ‬଴ଵ can be
condition of machines during an operation. Temperature written as,
signatures of the rotating machine can give an early warning
to the technician or operator to make a crucial decision before ܻ௟௢௪ ሺ‫݌‬ǡ ‫ݍ‬ሻ ൌ  ൣ‫ܮ‬௬ ‫ܫ  כ‬௢ ൧ ՝ଶ ሺ‫݌‬ǡ ‫ݍ‬ሻ (5)
occurrence of any critical problem in process or plant. ଵ
Few approaches [13-19] which makes use of the ൌ  ෍ ‫ܫ‬௢ ሺ‫݌‬ǡ ݇ሻ‫ܮ‬௬ ሺ‫݌‬ǡ ʹ‫ ݍ‬െ ݇ሻǡ
temperature features of thermal imaging to study the status of ௞ୀିଶ
rotating machines are available in the literature. Among
these approaches, thermal imaging for fault detection and ‫ݓ‬଴ଵ ሺ‫݌‬ǡ ‫ݍ‬ሻ ൌ  ൣ‫ܮ‬௫ ‫ כ‬ൣ‫ܮ‬௬ ‫ܫ  כ‬௢ ൧ ՝ଶ ൧ ՝ଶ ሺ‫݌‬ǡ ‫ݍ‬ሻ(6)
diagnosis of electrical rotating machineries have been ଵ
effectivelypresented in the literature [20-23] such as
misalignment, bearing, mass unbalance. Deep learning ൌ  ෍ ܻ௟௢௪ ሺ݇ǡ ‫ݍ‬ሻ‫ܮ‬௫ ሺʹ‫ ݌‬െ ݇ǡ ‫ݍ‬ሻǡ
network is applied for diagnosis of rolling or roll bearing ௞ୀିଶ
related fault [24] Vibration signature has good capability with Similar steps are performed to obtain ‫ݓ‬ଵଵ ǡ ‫ݓ‬ଶଵ ƒ†‫ݓ‬ଷଵ
the tunable Q-wavelet transform to classify defects through
various machine learning algorithms [25]. A non-contact
method has been developed for the bearing fault diagnosis
systems, based on frequency domain features of the vibration
data [26]. TIT is applied in the detection of abnormality
caused by overheating of stator windings [27-28]. A novel
method based on IR images Method of Areas Selection of
Image Differences (MoASoID) has been developed for
diagnosis of the broken bars and defected ring of squirrel-
cage rotor in induction motor [29-30]. A self adaptive
approach for fault diagnosis of cooling radiator based on six
types of radiator conditionhas been investigated with higher
classification accuracy using TIT [31]. Finally, the latest
Fig. 2. Two-level 2D DWT (a) Sub-bands functional diagram (b) Wavelet
study investigated for the self-adaptive roller bearingfault sub-bands block diagram
diagnosis system based on thermal images [32].
III. PROPOSED METHOD The three coefficient sub-image (HL, LH, HH) and low-
resolution sub-image (LL)as shown in Fig.3. A detailed
In this section, the proposed methodology for bearing literature about2D-DWThas been presented in [31].
fault detection and classification using 2D-DWT,
Mahalanobis distance, and SVM Classifier. Fig. 2 presents B. Feature Extraction
the flow chart of the proposed methodology. Image Feature may be determined and extracted by a
number of different approaches. These featuresmay include
A. Discrete Wavelet Transform
pixel information, boundary, region, and texture information.
The DWT has high performance and attains good ranks In this work, ten feature are extracted on the 2D
among the best decomposition techniques. It can perform approximation coefficient image computed according to
multi-resolution analysis in both frequency and timedomain. DWT [20]. The ten features are correlation, contrast, entropy,
The second decomposition level of the 2D-DWT was energy, kurtosis, homogeneity, RMS,mean, skewness and
applied on the thermal images of induction motor. The four Standard Deviation (STD). They are extracted from thermal
types of wavelet coefficients namely (vertical coefficients, imagesof the data set for four different faulty and healthy
horizontal coefficients or detailed components, diagonal and conditions of bearing in three phase induction motor. In each
approximation coefficients ) are obtained from each image. condition, there are twenty four samples of the thermal
The 2D-DWT decompose an image at each level providing images acquired in each condition of bearing.
four sub images, three coefficient sub-image (HL, LH, HH)
and low-resolution sub-image (LL). The scale of the original C. Feature Selection
thermal image (Io) at n= 0 can be specified by 2n = 20 = 1. Feature selection is different from dimension reduction.
Feature selection methods include and exclude attributes
Furthermore the sub-images in the result at n = 1 can be
present in the image data set without changing them. Itis the
written as;
distance between two points in multivariate space. In
‫ݓ‬଴ଵ ሺ‫݌‬ǡ ‫ݍ‬ሻ ൌ  ൣ‫ܮ‬௫ ‫ כ‬ൣ‫ܮ‬௬ ‫ܫ  כ‬௢ ൧ ՝ଶ ൧ ՝ଶ ሺ‫݌‬ǡ ‫ݍ‬ሻ (1) multivariate statistics, it is one of the most common
measureswhich can be used to determine whether a data
‫ݓ‬ଵଵ ሺ‫݌‬ǡ ‫ݍ‬ሻ ൌ  ൣ‫ܮ‬௫ ‫ כ‬ൣ‫ܪ‬௬ ‫ܫ  כ‬௢ ൧ ՝ଶ ൧ ՝ଶ ሺ‫݌‬ǡ ‫ݍ‬ሻ (2) sample is a part of a class or not. The Indian statistician P. C.
Mahanobis was first proposed Mahalanobis distance in 1936
‫ݓ‬ଶଵ ሺ‫݌‬ǡ ‫ݍ‬ሻ ൌ  ൣ‫ܪ‬௫ ‫ כ‬ൣ‫ܮ‬௬ ‫ܫ  כ‬௢ ൧ ՝ଶ ൧ ՝ଶ ሺ‫݌‬ǡ ‫ݍ‬ሻ (3) [32]. It is given as:
‫ݓ‬ଷଵ ሺ‫݌‬ǡ ‫ݍ‬ሻ ൌ  ൣ‫ܪ‬௫ ‫ כ‬ൣ‫ܪ‬௬ ‫ܫ  כ‬௢ ൧ ՝ଶ ൧ ՝ଶ ሺ‫݌‬ǡ ‫ݍ‬ሻ(4) d= [(x-y) C^(-1) (x-y)] (7)
where,
where,

951
x, y: points from the same distribution then draw an optimal hyperplane to dividedinto different
bearing defects. In this work we use SVMs for multi-class
C: covariance matrix classification, four cases of bearing condition, healthy inner
d : Mahalanobis distance race, outer race and lack of lubrication.
The distances amidst all the selected features
arecomputed using Mahalanobis distance. Hither, distance IV. EXPERIMENTAL STUDY
between the three categories, healthy versus inner race Thermal images of induction motor were acquired using
condition, healthy versus outer race condition and healthy FLIR E-60 (thermal camera) in an electrical laboratory at
versus lubrication problem. After finding Mahalanobis NITTTR, Chandigarh. The camera was installed at a distance
distance, the average distances of all different conditions are of 2 feet from the induction motor during the experimental
calculated and six strongest features are selected. The six setup as shown in Fig. 5. In this paper, thermal images of
features are mean, skewness, entropy, STD, energy, kurtosis. inductuion motor are captured under laboratory conditions.
Furthermore, the unnecessary features have been neglected, The specification of used induction motor is discussed in
conforming to the limit of relevance level. Table 1.

Fig. 4. Thermal image in four type of bearing conditions (a) Lack of


lubrication (b) Inner race defect (c) Outer race defect (d) Healthy

The thermal images of the four condition of bearing in the


three phase squirrel-cage induction motor: outer race, inner
race, lake of lubrication faults and healthy were acquired, in
each condition 24 thermal images are captured. The same
experiment is performed under the load conditions of the
motor full load,half-load and no-load. Therefore the total
number of thermal images are 288 (24*4*3). The
performance of the proposed approached is validate for
individual loading conditions, i.e each data set consider of 96
images which is divided into two group, training and testing
data sets. Fig.4 and Fig.5 illustrates the thermal images of
different faulty condition of bearings and experimental setup
respectively.

Fig. 3. Flow chart of proposed system


V. EXPERIMENTAL RESULT
At the ambient temperature, the induction motor is
D. Support Vector Machines switched ON. Its thermal signature will start to increase from
The self – adaptive fault diagnosis system has to its ambient temperature to a higher valuesteady state. In this
automaticallydiscriminate between different work it was observered tht after two hours of motor
bearingdefectcondition without human innervation for result operation steady state temperature was achieved. Thermal
interpretation. SVM is a supervised learning algorithm that images measured from the induction motor at three loading
can be used for both regression challenges and condition (no load, half load, full load ) and four faulty
classification.SVM is shown to be more effective for thermal conditions (outer race, inner race, lubricant fault and healthy,)
imaging based fault diagnosis than other learning techniques were firstly decomposed the themal images using 2D-DWT,
due to risk depreciation with higher accuracy for rotating after that desired wavelet coefficients are selected which is
machines [33-35]. The main objective of SVM is to make a used as input forthe stagethe feature extraction.
multi-dimensional hyperspace with the selected features and

952
TABLE I. SPECIFICATION OF INDUCTION MOTOR steady state temperature can be seen in Fig.6. Thermal
AND THERMAL CAMERA signature of healthy motor is lower than the faulty conditions.
Machine Machine
The rate of change of thermal signature is more in the case of
/Equipment
Specification
/Equipment lubricant, outer and inner race defect that of healthy bearing.
Induction Power Supply : 3 Phase 440V, Induction However, the difference between than the healthy and the
Motor 0.75A, 50Hz Motor three faulty bearing condition the bearing thermal signature
Power-: 0.5 HP increase with time. The fault size of outer and inner race
Speed: 1440 rpm defect is a 2mm hole as shown in Fig.1. In this paper, after
Thermal spectral range: 7.5–13 μm ( Thermal decomposition, six strongestfeatures were selected based on
Camera Uncooled microbolometer) Camera Mahalanobis distance methods.the best six features mean,
pixels : 320 × 240 skewness, entropy, STD, energy, kurtosis,are selected for
temperature range of Storage: passing onto the classifier. Figure 7 shows the results of the
– 40°C to +70°C classifier in terms of confusion matrix.It can be seen that the
temperature range (Operating) developed method is able to classify the four bearing
–15°C to +50°C conditions accurately.There is only one case of false
Relative humidity : +25°C to classification of lubrication fault as inner race fault in casee
+40°C / 2 cycles of no-load and half load conditions. In all the scenarios,
faulty conditions were identified from healthy bearning
This study, two levels of decomposition are applied to the condition. Therefore, it can be stated that the proposed
thermal images of the induction motor. Indeed, the intelligent thermal imaging based fault diagnosis and
approximation coefficient was considered since the most monitoring system is able to classify and diagnosticate the
prominent information contain in a low frequency signal in thermal images of three faulty and one healthy condition of
the original thermal image.The thermal signature of the bearing with a higher classification accuracy on the
proposed bearing conditions during 120 min to achieve its experimental dataset.

Fig. 5. Experimental Setup (1) FLIR E-60 (Thermal Camera), (2) Digital Multimeter (3) 3 Phase Power Supply (4) Direct On Line Starter (5) Digital
Tachometer (6) Induction Motor with Bearing Faults (7) Computer GUI

953
Fig. 6. Thermal signature of different bearing conditions (a) at no load (b) at half load (c) at full load

Fig. 7. Confusion matrix of SVM (a) at no-load (b) at half -load (c) at full-load

[4] P. Jayaswal, A. K. Wadhwani, and K. B. Mulchandani, “Machine fault


VI. CONCLUSION signature analysis,” Int. J. Rotating Mach., vol. 2008, pp. 1–10, Jan.
In this paper, authors develop an approach for intelligent 2008.
or self-adaptive bearingfault diagnosis system for the [5] Maldague X ,“Theory and Practice of Infrared Technology for Non
Destructive Testing” in John-Wiley & Sons, New Jersey,ISBN: 978-0-
induction motor using thermal imaging techniques. The 471-18190-3, USA 2001.
proposed system is based on the thermal images of induction
[6] R. Ishimwe, K. Abutaleb, and F. Ahmed, “Applications of thermal
motor in different loading and faulty conditions. The authors imaging in agriculture a review,” Advances in Remote Sensing, vol. 3,
analyzed the thermal images of four states of bearing in the no. 03, p. 128, 2014.
induction motor: healthy bearing, outer and inner race [7] A. Akula, R. Ghosh, and H. Sardana, “Thermal imaging and its
faultsand lubricant problem in the three phase induction application in defence systems,” in AIP Conference Proceeding,
motor. The results seems promising and need to be validated 2011.vol. 1391, pp. 333–335.
on a larger data set so that the proposed intelligent fault [8] J. Rinker, “Airborne infrared thermal detection of caves and
diagnosis system can be implemented as a condition crevasses,” Photogrammetric engineering and remote sensing, vol. 44,
no. 11, 1975.
monitoring tool for other rotating machines.
[9] M.-H. Wang, P.-C. Wu, and W.-J. Jiang, “Application of infrared
thermography and extension recognize method to intelligent fault
diagnosis of distribution panels,” IEEJ Transactions on Electrical and
REFERENCES Electronic Engineering, vol. 10, no. 4, pp. 479–486, 2015.
[1] D. Wang, W. T. Peter, and K. L. Tsui, “An enhanced Kurtogram
method for fault diagnosis of rolling element bearings,” Mech. Syst. [10] E. Ring and K. Ammer, “Infrared thermal imaging in medicine,”
Signal Process, vol. 35, pp. 176–99, Feb, 2013. Physiological measurement, vol. 33, no. 3, p. R33, 2012.
[2] J. Zarei, M. A. Tajeddine, and H. R. Karimi, “Vibration analysis for [11] R. Sonan, S. Harmand, J. Pell´e, D. Leger, and M. Fakes,“Transient
bearing fault detection and classification using an intelligent filter,” thermal and hydrodynamic model of flat heat pipe for the cooling of
Mechatronics, vol. 24, pp. 151–7, Mar. 2014. electronics components,” International Journal of Heat and Mass
Transfer, vol. 51, no. 25, pp. 60, 2015.
[3] H. Saruhan, S. Sandemir, A. Cicek, and I. Uygur, “Vibration analysis
of rolling element bearings defects,” J. Appl. Res. Technol., vol. 12, [12] Wontae KIM, Jinju SEO, Dongpyp H “Infrared thermographic
pp. 384–95, Jun. 2014. inspection of ball bearing; condition monitoring”. in 18th World
Conference of Non Destructive Testing, Durban, South Africa, 2012.

954
[13] Fantidis J. Gs., Karakoulidis K., Lazidis G., Potolias C., Bandekas D. diagnosis for rotating machinery,” Infrared Physics & Technology,
V., “The study of the thermal profile of a three-phase motor under vol. 73, pp. 78–87, 2015.
different conditions”, ARPN Journal of Engineering and Applied [24] Haidong Shao, Hongkai Jiang, Xun Zhang and Maogui Niu “Rolling
Sciences, 8 (11): 892 – 899 2013. bearing fault diagnosis using an optimization deep belief network”
[14] Garcia-Ramirez A, Morales-Hernandez L, Osornio-Rios R, Garcia- Measurement Science and Technology, vol. 26, no. 11, 2015.
Perez A, Romero-Troncoso“Thermographic technique as a [25] Upadhyay, Nitin, and Pavan Kumar Kankar. "Diagnosis of bearing
complement for mcsa in induction motor fault detection”. In: defects using tunable Q-wavelet transform." Journal of Mechanical
International conference on electrical machines (ICEM), pp 1940– Science and Technologyvol. 32, pp 549-558, 2018.
1945, 2014.
[26] Arun, P., S. Abraham Lincon, and N. Prabhakaran. "Detection and
[15] Armando G. Garcia-Ramirez, Luis A. Morales-Hernandez, Roque A. Characterization of Bearing Faults from the Frequency Domain
Osornio-Rios, Juan P. Benitez-Rangel, Arturo Garcia-Perez, Rene de Features of Vibration." IETE Journal of Research: pp. 1-14, 2017.
J. Romero-Troncoso, “Fault detection in induction motors and the
impact on the kinematic chain through thermographic analysis”, [27] M. Eftekhari, M. Moallem, S. Sadri, H. Min-Fu, “A novel indicator of
Electr. Power Syst. Res. 114, pp1–9, 2014. stator winding fault in induction motor using infrared thermal
imaging”, Infrared Phys. Technol.Vol. 61, pp. 330–336, 2013.
[16] D.K. Chaturvedi, M.S. Iqbal, M.P. Singh, “Intelligent health
[28] G. T. Singh, C. Anil Kumar, and V. N. A. Naikan, “Induction motor
monitoring system for three phase induction motor using infrared
inter turn fault detection using infrared thermographic analysis,”
thermal image”, in International Conference on Energy Economics
Infrared Phys. Technol., vol. 77, pp. 277–282, 2016.
and Environment (ICEEE), Greater Noida, India, 2015.
[29] A. Glowacz and Z. Glowacz, “Diagnostics of stator faults of the
[17] J.A.R. Nunez, L.M. Velazquez, L.A.M. Hernandez, R.J.R. Troncoso,
single- phase induction motor using thermal images, moasos and
R.A. Osornio-Rios, “Low-cost thermographic analysis for bearing
selected classifiers,” Measurement, vol. 93, pp. 86–93, 2016.
fault detection on induction motors”, J. Sci. Ind. Res. 75 (7) 412–415,
2016. [30] A. Glowacz and Z. Glowacz, “Diagnosis of the three- phase induction
[18] Gurmeet Singh, T. Ch. Anil Kumar, V.N.A. Naikan “Fault diagnosis motor using thermal imaging,” Infrared Physics & Technology, vol.
81, pp. 7–16, 2017.
of induction motor cooling system using infrared thermography” in
6th IEEE International Conference on Power Systems (ICPS), New [31] A. Taheri-Garavand, H. Ahmadi, M. Omid, S. S. Mohtasebi, K. Mol-
Delhi, India, 2016. lazade, A. J. R. Smith, and G. M. Carlomagno, “An intelligent
approach for cooling radiator fault diagnosis based on infrared thermal
[19] Wael Moussa “A passive thermography approach to bearing condition
image processing technique,” Applied Thermal Engineering, vol. 87,
monitoring” in Juniper Online Journal Material Science, vol. 1 Iss. 4 -
pp. 434– 443, 2015.
May 2017.
[32] Huo, Z.; Zhang, Y.; Sath, R.; Shu, L. “Self-adaptive fault diagnosis of
[20] A. M. Younus and B.-S. Yang, “Intelligent fault diagnosis of rotating
roller bearings using infrared thermal images”, In Proceedings of the
machinery using infrared thermal image,” Expert Systems with
Applications, vol. 39, no. 2, pp. 2082–2091, 2012. 43rd Annual Conference of the IEEE Industrial Electronics Society
(IECON 2017), Beijing, China, 29 October–1 November 2017.
[21] F. Jeffali, B. EL Kihel, A. Nougaoui, and F. Delaunois, “Monitoring
and diagnostic misalignment of asynchronous machines by infrared [33] A. Widodo and B.-S. Yang, “Support vector machine in machine
thermography,” J. Mater. Environ. Sci., vol. 6, no. 4, pp. 1192–1199, condition monitoring and fault diagnosis,” Mechanical systems and
2015. signal processing, vol. 21, no. 6, pp. 2560–2574, 2007.
[34] V. Vapnik, “The nature of Statistical Learning Theory”, New York:
[22] L. Duan, M. Yao, J. Wang, T. Bai, and L. Zhang, “Segmented infrared
Wiley,Second Edition, ISBN:0-387-94559-8 1998.
image analysis for rotating machinery fault diagnosis”, Infrared
Physics & Technology, vol. 77, pp. 267–276, 2016. [35] Khemchandani R, Chandra S, “Twin support vector machines for
pattern classification” IEEE Transactions on pattern analysis and
[23] O. Janssens, R. Schulz, V. Slavkovikj, K. Stockman, M. Loccufier, R.
machine intelligencevol. 29(5), pp. 905–910, 2007.
Van de Walle, and S. Van Hoecke, “Thermal image based fault

955

You might also like