The Report is Generated by DrillBit Plagiarism Detection Software

Submission Information

Author Name H S Kumar

Title Classification
Paper/Submission ID 2791349
Submitted by urkumar2006@nitte.edu.in
Submission Date 2024-12-17 14:12:24
Total Pages, Total Words 8, 2548
Document type Research Paper

Result Information

Similarity 18 %
1 10 20 30 40 50 60 70 80 90

Sources Type Report Content

Student
Words <
Paper
14,
0.75% Internet
6.16%
5.38%

Journal/
Publicatio Ref/Bib
n 11.88% 28.38%

Exclude Information Database Selection

Quotes Not Excluded Language English

References/Bibliography Excluded Student Papers Yes
Source: Excluded < 14 Words Not Excluded Journals & publishers Yes
Excluded Source 0% Internet or Web Yes
Excluded Phrases Not Excluded Institution Repository Yes

A Unique QR Code use to View/Download/Share Pdf File

DrillBit Similarity Report

A-Satisfactory (0-10%)
B-Upgrade (11-40%)

18 29 B C-Poor (41-60%)
D-Unacceptable (61-100%)
SIMILARITY % MATCHED SOURCES GRADE

LOCATION MATCHED DOMAIN % SOURCE TYPE

1 Condition Monitoring with Vibration Signals (Compressive Sampling Publication

1
and

2 extrica.com Internet Data

<1

3 Classification of ictal and seizure-free EEG signals using fractional, by Publication

1
Joshi, Varun Pacho- 2014

5 www.iieta.org Publication
1

6 nmamit.nitte.edu.in Internet Data

1

7 extrica.com Publication
1

9 extrica.com Internet Data

<1

10 ouci.dntb.gov.ua Internet Data

<1

13 extrica.com Internet Data

1

14 Deep convolutional neural network based planet bearing fault Publication

1
classification by Zhao-2019

15 Thesis Submitted to Shodhganga Repository Publication

1

17 Submitted to Visvesvaraya Technological University, Belagavi Student Paper

1
18 IEEE 2017 14th IAPR International Conference on Document Analysis Publication
1
a, by Venugopal, Vivek S- 2017

22 Thesis submitted to shodhganga - shodhganga.inflibnet.ac.in Publication

1

23 www.dx.doi.org Publication
1

26 worldwidescience.org Internet Data

1

27 1library.co Internet Data

<1

28 Thesis Submitted to Shodhganga Repository Publication

<1

29 www.sciencegate.app Internet Data

<1

30 A methodology for identification of damage in beams by Barbieri-2015 Publication

<1

31 jddtonline.info Publication
<1

32 Thesis submitted to shodhganga - shodhganga.inflibnet.ac.in Publication

<1

33 A Physics-Based Modeling Approach for Performance Monitoring in Gas Publication

<1
Tu by Hanachi-2015

34 bmcbioinformatics.biomedcentral.com Internet Data

<1

35 Electrotonic Coupling in the Pituitary Supports the Hypothalamic- Publication

<1
Pituitary-Gonad by Gngrich-2016

36 extrica.com Internet Data

<1

37 moam.info Internet Data

<1

38 www.ijecce.org Publication
<1

39 www.ijrter.com Publication
<1
Classification of Vibration
30
Signal Using Wavelet Transform
and Support
29
Vector Machine for Rolling Element Bearing
Fault diagnosis
Kumar H S1

1Associate
6
Professor, Department of Mechanical Engineering, NMAM Institute of Technology, NITTE (Deemed
to be University), Nitte, Karnataka, 574110, India

7 10
Abstract- This study focusses on fault diagnosis in Rolling Element Bearings (REB) using Discrete Wavelet
transform (DWT), Statistical Features (SFs) and Support Vector Machine (SVM). For effective classification, SFs
are extracted from the second level of DWT wavelet coefficients at fourth level. These features were fed to SVM
classifier to classify the conditions of REB. The experimental result shows that, REB conditions can be classified
effectively using SVM.

Keywords: Vibration signal, DWT, statistical features, SVM

Introduction
REB are vital components in rotating machinery, and their health directly impacts operational
efficiency and safety. Commonly occurring bearing faults like galling, spalling, peeling and
failure of the bearing due to misalignment, shaft surface roughness and higher waviness and
inclusions etc. causes sudden failure. Unexpected failures can lead to significant downtime and
14
financial losses. Therefore, effective fault diagnosis methods are essential for predictive
maintenance and ensuring reliable operation [1-2]. Vibration analysis is a crucial technique
for machinery fault diagnosis due to its strong correlation with bearing health conditions. It is
2
different from the other methods, as reported in [3-5, 8-10].

Time, Frequency and Time-Frequency (T-F) methods are widely used in vibration signal
analysis. Time domain approach analyses signals based on their variation over time and is
ineffective for non-stationary signals as it does not provide frequency information. Frequency
domain methods transform signals into the frequency space using techniques like the Fourier
Transform (FT) and loses temporal information, making it difficult to analyse non-stationary
signals. As reported in literature [9], T-F features are effective in analysing non-stationary
vibration signals obtained from machines [9-10]. Thus, T-F approach is used in this work.
Additionally, wavelet analysis uses T-F information at a time, hence it is superior as compared
to time and spectrum techniques [2]. The Wavelet Transform (WT) is utilized to capture
various transient characteristics in vibration signals caused by bearing faults. WT can be
categorized into three types: Continuous Wavelet Transform (CWT), DWT and Wavelet
Packet Transform (WPT). Various researches used different TF techniques for REB condition
classification as stated in the literature [2-8]. Prabhakar et al. (2002) used DWT for fault
diagnosis of REB. Saruhan et al. [13] performed vibration analysis on REB with defects in IR,
17
OR, and ball, as well as a combination of these defects, under various load and speed
conditions. Jena and Panigrahi [14] measured the width of defects introduced on the IR and
OR of the REB vibration signal acquired from the test setup developed by them.
Fault diagnosis comprises of three stages namely fault detection, isolation and classification
14
which may involves data derived from the vibration signals of the rotary machinery [1].
Researchers proposed different fault diagnosis schemes as follows: Li & Zhao proposed how
the integration of WPT and SVM offers an effective approach for REB fault diagnosis. The
energy feature vector derived from the decomposed WPT coefficients varies as the condition
of the REB changes. Thus, energy value of the different states of REB were fed to SVM
1
classifier which effectively classified the REB conditions [15]. Skariah et al. (2016) presented
33
a robust and effective method for health monitoring of REB by integrating an improved
Wavelet Cross Spectrum (WCS) technique with SVM for fault diagnosis. The WCS technique
obtains features associated with bearing health parameters by cross-correlating the periodic
5
patterns across the most relevant wavelet bands. The proposed method offers significant
potential for real-time predictive maintenance applications, helping to ensure the reliability and
longevity of critical machinery. They concluded that improved WCS resulted in satisfactory
1
results [16]. Yang et al. (2008) proposed a method for fault diagnosis in rolling bearings by
combining WPT for feature extraction and SVM for classification. He concluded that SVM is
superior in comparison with neural network whenever sample size is low.
36 28
In this work, SVM is introduced into fault diagnosis of REB using the vibration signals
acquired under healthy and faulty conditions i.e., N, IR and OR. The signals are acquired at
23
maximum load and at different speed. Seventeen SFs are obtained from the decomposed DWT
coefficients are fed to SVM to test the performance of the classifier. The structure of the paper
7
is as follows: section 1 contains introduction; brief history of wavelet transform is discussed in
13
section 2. Experimental setup and feature extraction is presented in section 3. Section 4
describes brief information about SVM classifier and in section 5 discusses the outcomes of
the work 5 followed by conclusions in section 6.
 2 Wavelet Transform

Continuous WT of a signal at time u and scale s is given in equ. (1)

The CWT of a signal is given by
1
퐶푊푇 (푢, 푠 ) = √ꢀ
∫푅 푓 (푡 ) 휑 ꢁꢂ−ꢃ
ꢀ
ꢄ 푑푡 (1)

where φ (t) is the mother wavelet u and s indicate time and scale parameter.
DWT is preferred in digital computer application [18]. In DWT, the signal will be split into an
approximation portion (cA1) and a detail portion (cD1) as shown in fig.1, using a low-pass
filter and a high-pass filter, as follows [8]:
ꢇ
cAꢅ = ∑ꢈ 퐿(푘 ꢆ 2푛)푆(푘) (2)
ꢇ
cDꢅ = ∑ꢈ 퐻(푘 ꢆ 2푛) 푆(푘) (3)

Fig.1 DWT decomposition step [8].

The equation for DWT is given by 4 which is obtained from the discretisation of CWT (u, s)

1 ꢋ
+∞ ꢂ−ꢊ ꢈ
퐷푊푇(푗, 푘) =
ꢉ ꢊꢋ
∫−∞ ƒ(푡) 휓 ꢁ ꢊꢋ ꢄ (4)

where u and s are swapped by 2j and 2j k. As reported in [8], cD2 coefficient has maximum

energy and hence seventeen SFs have been extracted from the cD2 coefficient.
 2 Experimental test rig and feature extraction

Fig.2 shows the test rig to acquire the REB signals for healthy and different defective bearings
in this work. More detailed explanation of the REB test rig is given in [17-18], which provides
31
information about the REB test rig details and the software used for data acquisition (DAQ).
5
Table 1 Details of REB data set used in this work.
Bearing type Conditions Maximum load Variable Sampling Accelerometers Data
of REB speed rate size
Deep groove N 356 rpm X 250000
ball bearing IR 1.7 kN 622 rpm 48000 Y
OR

Fig.2 Schematic diagram of REB test rig.

26
The dataset used in this work is used to assess the accuracy of the SVM classifier to identify
the three conditions of the REB i.e., healthy and two defective conditions of the REB. Signals
were collected for three states of REB with maximum load, variable speed and sampling
35
frequency as mentioned in table1 for a duration of 5.08s. Signals were acquired using two
26
accelerometers (X and Y) and signals pertaining to X is considered for further analysis as they
are more sensitive when compared to Y. This resulted in data matrix of 250000 x 1 for one
18
load and two speed. Hence, totally six such data vectors were formed [8,18]. Each data vector
is split into 25 segments with each segment comprising of 10000 samples. The details of the
SFs used in this paper are given in [8] and are extracted from the cD2 coefficients which forms
a single pattern. Therefore, for 1.7 kN load and variable speed resulted in 150 patterns (25×6).
The feature values are normalized which has range of [0-1]. The patterns are randomly mixed
and out of which 70 % forms training data and the remaining 30 % forms test data.

3 SVM classifier

27
SVM is also commonly termed as support vector networks, are supervised learning models
37 3
which investigates the data for classification and regression analysis in machine learning. It is
34
based on statistical approach which is one of the widely used prediction method. SVM models
have been developed firstly when data possess precisely two categories meaning the class out
9
put can be designated with A and B as shown in fig. 3. But real time problems posses more
than two categories. For example, REB condition classification. There could be many defective
states on the basis of location of defects on different components of REB such as IR, OR cage
defect, combination defect etc. Hence real-world problems are multiclass problem and hence
3
in this work SVM is used for REB condition classification and mathematical description about
SVM can be found in [20-23].

Fig. 3 Data classification by SVM [11].

If the data point is not separated linearly, such points can be separated by projecting into higher
dimension. This makes the data linearly separated and the decision boundary generated in this
extra dimension can be transferred back to the original lower dimension by using mathematical
transformation. This can be performed using kernel functions and literature review [11, 20-22]
reveals different kernel function can be used to compute classification accuracy of SVM.

4 Results and Discussions

A total of 70 % data i.e. (17 ×105) is used for training the SVM classifier. The classification
38 32
learner app in MATLAB 2024 is used to train the SVM classifier [ ]. Result of the trained SVM
15
classifier along with confusion matrix for 3-fold cross validation with cubic kernel function
shown in fig.5. The confusion matrix presents the trained data in the form of correctly classified
and incorrectly classified data [22-23]. The classification accuracy on the trained data resulted
in 97.1 % training accuracy. The SVM classifier is trained using linear, quadratic, cubic and
fine, medium and coarse gaussian function. Among these, SVM model with cubic kernel
function resulted in highest training accuracy as shown in fig.4.
5
The remaining 30 % of the data i.e. (17 × 45) is tested using the trained SVM model and the
confusion matrix of the tetsed SVM is shown in fig.5. In the confusion matrix diagonal
elements represents the correctly classified whereas non-diagonal elements as not correctly
1
classified. The classification accuracy on the test data resulted in 93.33 % on the test data.
Confusion matrix reveals that N condition resulted in 100% classification accuracy on test data
followed by 93.33% and 86.66% on IR and OR. Therefore, overall classification accuracy of
the SVM classifier resulted in 93.33 % accuracy on test data.

Fig.4 Confusion matrix of the trained model of SVM.

 Fig.5 Confusion matrix of the tested SVM classifier.

7 Conclusions

This paper proposes a methodology to classify REB conditions using SFs extracted using
DWT and multiclass SVM classifier. Seventeen SFs have been extracted from second level
22
cD2 DWT wavelet coefficients. The extracted SFs are fed to SVM classifier and it is trained
using different kernel functions which resulted in 97.1 % classification accuracy on training
data using cubic kernel function. The trained SVM classifier resulted in 93.33 % overall
classification accuracy on the test data. However, classification accuracy can be improved by
using the denoised signal, different mother wavelet and optimizing the features considered in
39
this work. The results reveal that proposed method can effectively classify the conditions of
REB

References

1. Rai, A., & Upadhyay, S. H. (2016). A review on signal processing techniques utilized in the fault
diagnosis of rolling element bearings. Tribology International, 96, 289-306.
2. Zhang, X., Liu, Z., Wang, J., & Wang, J. (2019). Time–frequency analysis for bearing fault diagnosis
using multiple Q-factor Gabor wavelets. ISA transactions, 87, 225-234.
3. Ebersbach, S., Peng, Z., & Kessissoglou, N. J. (2006). The investigation of the condition and faults of a
spur gearbox using vibration and wear debris analysis techniques. Wear, 260(1-2), 16-24.
4. Yang, H., Jiang, H., Yang, Z., & Li, J. (2009, April). Fault diagnosis of rolling bearing based on wavelet
packet transform and support vector machine. In 2009 International Conference on Measuring
Technology and Mechatronics Automation (Vol. 1, pp. 650-653). IEEE.
5. Sharifi, S., Tivay, A., Rezaei, S. M., Zareinejad, M., & Mollaei-Dariani, B. (2018). Leakage fault
detection in Electro-Hydraulic Servo Systems using a nonlinear representation learning approach. ISA
transactions, 73, 154-164.
6. Van Hecke, B., Yoon, J., & He, D. (2016). Low speed bearing fault diagnosis using acoustic emission
sensors. Applied Acoustics, 105, 35-44.
7. He, C., Li, H., & Zhao, X. (2018). Weak characteristic determination for blade crack of centrifugal
compressors based on underdetermined blind source separation. Measurement, 128, 545-557.
8. Zhang, X., Miao, Q., Liu, Z., & He, Z. (2017). An adaptive stochastic resonance method based on grey
wolf optimizer algorithm and its application to machinery fault diagnosis. ISA transactions, 71, 206-214.
9. Kumar, H. S., Pai, P. S., Sriram, N. S., & Vijay, G. S. (2013). ANN based evaluation of performance of
wavelet transform for condition monitoring of rolling element bearing. Procedia engineering, 64, 805-
814.
10. Akan, A., & Cura, O. K. (2021). Time–frequency signal processing: Today and future. Digital Signal
Processing, 119, 103216.
11. Jha, R. K., & Swami, P. D. (2021). Fault diagnosis and severity analysis of rolling bearings using
vibration image texture enhancement and multiclass support vector machines. Applied Acoustics, 182,
108243.
12. Peng, Z. K., & Chu, F. L. (2004). Application of the wavelet transform in machine condition monitoring
and fault diagnostics: a review with bibliography. Mechanical systems and signal processing, 18(2), 199-
221.
13. Prabhakar, S., Mohanty, A. R., & Sekhar, A. S. (2002). Application of discrete wavelet transform for
detection of ball bearing race faults. Tribology International, 35(12), 793-800.
14. Saruhan, H., Saridemir, S., Qicek, A., & Uygur, I. (2014). Vibration analysis of rolling element bearings
defects. Journal of applied research and technology, 12(3), 384-395.
15. Jena, D. P., & Panigrahi, S. N. (2014). Precise measurement of defect width in tapered roller bearing
using vibration signal. Measurement, 55, 39-50.
16. Li, M., & Zhao, P. (2008, August). The application of wavelet packet and svm in rolling bearing fault
diagnosis. In 2008 IEEE International Conference on Mechatronics and Automation (pp. 505-508).
IEEE.
17. Skariah, A., Pradeep, R., Rejith, R., & Bijudas, C. R. (2021). Health monitoring of rolling element
bearings using improved wavelet cross spectrum technique and support vector machines. Tribology
International, 154, 106650.
18. Kumar, H. S., & Manjunath, S. H. (2022). Use of empirical mode decomposition and K-nearest
neighbour classifier for rolling element bearing fault diagnosis. Materials Today: Proceedings, 52, 796-
801.
19. Kumar, H. S., Gururaj, U., & Srinivasa, P. P. (2021). Rolling element bearing health diagnosis–A
comparative study of statistical features. Materials Today: Proceedings, 46, 2666-2672.
20. Yang, H., Jiang, H., Yang, Z., & Li, J. (2009, April). Fault diagnosis of rolling bearing based on wavelet
packet transform and support vector machine. In 2009 International Conference on Measuring
Technology and Mechatronics Automation (Vol. 1, pp. 650-653). IEEE.
21. Thelaidjia, T., Moussaoui, A., & Chenikher, S. (2015, December). Bearing fault diagnosis based on
independent component analysis and optimized support vector machine. In 2015 7th International
Conference on Modelling, Identification and Control (ICMIC) (pp. 1-4). IEEE.
22. Yuan, L., Lian, D., Kang, X., Chen, Y., & Zhai, K. (2020). Rolling bearing fault diagnosis based on
convolutional neural network and support vector machine. IEEE Access, 8, 137395-137406.
23. Lv, Z., Tang, B., Zhou, Y., & Zhou, C. (2016). A novel method for mechanical fault diagnosis based on
variational mode decomposition and multikernel support vector machine. Shock and Vibration, 2016(1),
3196465.
24. Agrawal, P., & Jayaswal, P. (2020). Diagnosis and classifications of bearing faults using artificial neural
network and support vector machine. Journal of The Institution of Engineers (India): Series C, 101(1),
61-72.