# Performance of Bearing Ball Defect Classification Based on the Fusion of Selected Statistical Features

^{1}

^{2}

^{3}

^{4}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Paper Contribution

## 3. The Fault Diagnosis Methodology

#### 3.1. Preprocessing and Feature Extraction and Selection

- The mean and the skewness had very poor detection performance
- The kurtosis had a very low sensitivity to the ball fault level.

**Variance**: $IM{F}_{2}$, $IM{F}_{3}$ and $IM{F}_{4}$; denoted as $vIM{F}_{j}$**KLD**: $IM{F}_{2}$, $IM{F}_{3}$, $IM{F}_{4}$ and $IM{F}_{6}$; denoted as $kIM{F}_{j}$.

#### 3.2. Feature Analysis

#### 3.2.1. Principal Component Analysis (PCA)

- Kernel-based techniques: kernel principal component analysis (KPCA) and support vector machine (SVM);
- Deterministic systematic exploration techniques: K-nearest neighbours (KNN) and decision tree (DT);
- Probabilistic systematic exploration techniques: naive Bayes classifiers (NB).

#### 3.2.2. Kernel Principal Component Analysis (KPCA)

- The polynomial kernel defined as ( $p\in {\mathbb{R}}^{+}$ is the kernel’s order):$$K(x,y)={\left(\langle x,y\rangle +1\right)}^{p}$$
- The Gaussian kernel defined as ( $\gamma \in {\mathbb{R}}^{+}$ is the standard deviation of the kernel):$$K(x,y)=exp\left(-\frac{{\parallel x-y\parallel}^{2}}{2{\gamma}^{2}}\right)$$

#### 3.2.3. Support Vector Machine (SVM)

#### 3.2.4. K-Nearest Neighbours (KNN)

- Euclidean distance (Euc). It is defined by:$$Euc(A,B)=\sqrt{\sum _{i=1}^{m}{({x}_{i}-{y}_{i})}^{2}}$$
- City block distance (CB) given as:$$CB(A,B)=\sum _{i=1}^{m}|{x}_{i}-{y}_{i}|$$

#### 3.2.5. Decision Tree (DT)

#### 3.2.6. Naive Bayes (NB)

## 4. Results and Discussions

#### 4.1. Experimental Data

- No-load condition (${L}_{0}$): $0\%$ of the nominal load;
- Half-loaded condition (${L}_{1}$): $50\%$ of the nominal load;
- Fully loaded condition (${L}_{2}$): $100\%$ of the nominal load;
- Overloaded condition (${L}_{3}$): $150\%$ of the nominal load;
- Combination of all the load conditions: (${L}_{n}$).

- H: corresponding to the healthy behaviour (no fault);
- ${F}_{1}$: faulty case with a severity of $0.007$ inch;
- ${F}_{2}$: faulty case with a severity of $0.014$ inch;
- ${F}_{3}$: faulty case with a severity of $0.021$ inch.

#### 4.2. Experimental Validation

#### 4.2.1. Linear Classification with PCA

#### 4.2.2. Kernel-Based Classifiers

#### 4.2.3. Classification Results Based on the Systematic Data Exploration Strategy

- Under the single-load condition, all the three classifiers exhibited good performance despite a low testing accuracy rate of $96.5\%$ for the NB classifier;
- Under the combined-load condition, the performance of the NB classifier was severely degraded with $82.3\%$ and $81.92\%$ for the training accuracy rate and the testing accuracy rate, respectively.

- ■
- Case study with four features
- □
- KLD and variance of $IM{F}_{2}$ and $IM{F}_{4}$, denoted as $C4$;

In this case study, the KLD and the variance of the selected IMFs ($IM{F}_{2}$ and $IM{F}_{4}$) were merged together for each load condition as in the following matrix.$$\left\{\begin{array}{cccc}kIM{F}_{2{,}_{1}}& vIM{F}_{2{,}_{1}}& kIM{F}_{4{,}_{1}}& vIM{F}_{4{,}_{1}}\\ kIM{F}_{2{,}_{2}}& vIM{F}_{2{,}_{2}}& kIM{F}_{4{,}_{2}}& vIM{F}_{4{,}_{2}}\\ \vdots & \vdots & \vdots & \vdots \\ kIM{F}_{2{,}_{900}}& vIM{F}_{2{,}_{900}}& kIM{F}_{4{,}_{900}}& vIM{F}_{4{,}_{900}}\end{array}\right\}$$ - ■
- Case study with two features
- □
- KLD and variance of $IM{F}_{2}$, denoted as $C{2}_{1}$;
- □
- KLD and variance of $IM{F}_{4}$, denoted as $C{2}_{2}$;
- □
- Variances of $IM{F}_{2}$ and $IM{F}_{4}$, denoted as $C{2}_{3}$;
- □
- KLD of $IM{F}_{2}$ and $IM{F}_{4}$, denoted as $C{2}_{4}$.

- ■
- Case study with one feature
- □
- Variance of $IM{F}_{2}$, denoted as $C{1}_{1}$;
- □
- Variance of $IM{F}_{4}$, denoted as $C{1}_{2}$;
- □
- KLD of $IM{F}_{2}$, denoted as $C{1}_{3}$;
- □
- KLD of $IM{F}_{4}$, denoted as $C{1}_{4}$.

- For both training and testing steps, whatever the load condition or the used classifier, case $C4$ with the combination of KLD and variance for $IM{F}_{2}$ and $IM{F}_{4}$ offers the best performance.
- This analysis shows that it is possible to adapt to each case and meet the application requirements. Taking the example of load ${L}_{3}$, we can choose to work with either four features ($C4$), two features ($C{2}_{1}$) or even one feature ($C{1}_{1}$: variance of $IM{F}_{2}$) to reach $100\%$ of classification accuracy. This flexibility can address the computation time that is strongly linked to the number of used features corresponding to the input’s dimension of the classification system.
- Finally, we can conclude that in our study, the KNN classifier offers the most efficient combinations of features.

## 5. Conclusions

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Conflicts of Interest

## References

- Huang, W.; Cheng, J.; Yang, Y. Rolling bearing fault diagnosis and performance degradation assessment under variable operation conditions based on nuisance attribute projection. Mech. Syst. Signal Process.
**2019**, 114, 165–188. [Google Scholar] [CrossRef] - Wei, Y.; Li, Y.; Xu, M.; Huang, W. A Review of Early Fault Diagnosis Approaches and Their Applications in Rotating Machinery. Entropy
**2019**, 21, 409. [Google Scholar] [CrossRef] [PubMed] - Tang, M.; Liao, Y.; Luo, F.; Li, X. A Novel Method for Fault Diagnosis of Rotating Machinery. Entropy
**2022**, 24, 681. [Google Scholar] [CrossRef] [PubMed] - Zarei, J.; Tajeddini, M.A.; Karimi, H.R. Vibration analysis for bearing fault detection and classification using an intelligent filter. Mechatronics
**2014**, 24, 151–157. [Google Scholar] [CrossRef] - Immovilli, F.; Bianchini, C.; Cocconcelli, M.; Bellini, A.; Rubini, R. Bearing fault model for induction motor with externally induced vibration. IEEE Trans. Ind. Electron.
**2012**, 60, 3408–3418. [Google Scholar] [CrossRef] - Li, Y.; Wang, X.; Si, S.; Huang, S. Entropy based fault classification using the Case Western Reserve University data: A benchmark study. IEEE Trans. Reliab.
**2019**, 69, 754–756. [Google Scholar] [CrossRef] - Tahir, M.M.; Khan, A.Q.; Iqbal, N.; Hussain, A.; Badshah, S. Enhancing fault classification accuracy of ball bearing using central tendency based time domain features. IEEE Access
**2016**, 5, 72–83. [Google Scholar] [CrossRef] - Boudiaf, A.; Moussaoui, A.; Dahane, A.; Atoui, I. A comparative study of various methods of bearing faults diagnosis using the Case Western Reserve University data. J. Fail. Anal. Prev.
**2016**, 16, 271–284. [Google Scholar] [CrossRef] - Safizadeh, M.; Latifi, S. Using multi-sensor data fusion for vibration fault diagnosis of rolling element bearings by accelerometer and load cell. Inf. Fusion
**2014**, 18, 1–8. [Google Scholar] [CrossRef] - Kaya, Y.; Kuncan, M.; Kaplan, K.; Minaz, M.R.; Ertunç, H.M. A new feature extraction approach based on one dimensional gray level co-occurrence matrices for bearing fault classification. J. Exp. Theor. Artif. Intell.
**2021**, 33, 161–178. [Google Scholar] [CrossRef] - Delpha, C.; Diallo, D.; Harmouche, J.; Benbouzid, M.; Amirat, Y.; Elbouchickhi, E. Bearing Fault Diagnosis in Rotating Machines. In Electrical Systems 2: From Diagnosis to Prognosis; Soualhi, A., Razik, H., Eds.; ISTE: London, UK, 2020; pp. 123–152. [Google Scholar]
- Neupane, D.; Seok, J. Bearing Fault Detection and Diagnosis Using Case Western Reserve University Dataset With Deep Learning Approaches: A Review. IEEE Access
**2020**, 8, 93155–93178. [Google Scholar] [CrossRef] - Qi, J.; Gao, X.; Huang, N. Mechanical Fault Diagnosis of a High Voltage Circuit Breaker Based on High-Efficiency Time-Domain Feature Extraction with Entropy Features. Entropy
**2020**, 22, 478. [Google Scholar] [CrossRef] - Jardine, A.K.; Lin, D.; Banjevic, D. A review on machinery diagnostics and prognostics implementing condition-based maintenance. Mech. Syst. Signal Process.
**2006**, 20, 1483. [Google Scholar] [CrossRef] - Liu, M.K.; Weng, P.Y. Fault Diagnosis of Ball Bearing Elements: A Generic Procedure based on Time-Frequency Analysis. Meas. Sci. Rev.
**2019**, 19, 185–194. [Google Scholar] [CrossRef] - Du, C.; Gao, S.; Jia, N.; Kong, D.; Jiang, J.; Tian, G.; Su, Y.; Wang, Q.; Li, C. A High-Accuracy Least-Time-Domain Mixture Features Machine-Fault Diagnosis Based on Wireless Sensor Network. IEEE Syst. J.
**2020**, 14, 4101–4109. [Google Scholar] [CrossRef] - Griffin, D.; Lim, J. Signal estimation from modified short-time Fourier transform. IEEE Trans. Acoust. Speech Signal Process.
**1984**, 32, 236–243. [Google Scholar] [CrossRef] - Allen, J.B.; Rabiner, L.R. A unified approach to short-time Fourier analysis and synthesis. Proc. IEEE
**1977**, 65, 1558–1564. [Google Scholar] [CrossRef] - Daubechies, I. The wavelet transform, time-frequency localization and signal analysis. IEEE Trans. Inf. Theory
**1990**, 36, 961–1005. [Google Scholar] [CrossRef] - Shensa, M.J. The discrete wavelet transform: Wedding the a trous and Mallat algorithms. IEEE Trans. Signal Process.
**1992**, 40, 2464–2482. [Google Scholar] [CrossRef] - Yan, R.; Gao, R.X. Hilbert–Huang transform-based vibration signal analysis for machine health monitoring. IEEE Trans. Instrum. Meas.
**2006**, 55, 2320–2329. [Google Scholar] [CrossRef] - Susanto, A.; Liu, C.H.; Yamada, K.; Hwang, Y.R.; Tanaka, R.; Sekiya, K. Application of Hilbert–Huang transform for vibration signal analysis in end-milling. Precis. Eng.
**2018**, 53, 263–277. [Google Scholar] [CrossRef] - Luo, Z.; Liu, T.; Yan, S.; Qian, M. Revised empirical wavelet transform based on auto-regressive power spectrum and its application to the mode decomposition of deployable structure. J. Sound Vib.
**2018**, 431, 70–87. [Google Scholar] [CrossRef] - Amirat, Y.; Elbouchickhi, E.; Delpha, C.; Benbouzid, M.; Diallo, D. Modal Decomposition for Bearing Fault Detection. In Electrical Systems 1: From Diagnosis to Prognosis; Soualhi, A., Razik, H., Eds.; ISTE: London, UK, 2020; pp. 121–168. [Google Scholar]
- Shi, R.; Wang, B.; Wang, Z.; Liu, J.; Feng, X.; Dong, L. Research on Fault Diagnosis of Rolling Bearings Based on Variational Mode Decomposition Improved by the Niche Genetic Algorithm. Entropy
**2022**, 24, 825. [Google Scholar] [CrossRef] [PubMed] - Huang, N.E.; Shen, Z.; Long, S.; Wu, M.; Shih, H.; Zheng, Q.; Tung, C.; Liu, H. The empirical mode decomposition and Hilbert spectrum for nonlinear and nonstationary time series analysis. Proc. R. Soc. A
**1998**, 545, 903–995. [Google Scholar] [CrossRef] - Ahmed, H.O.A.; Nandi, A.K. Intrinsic Dimension Estimation-Based Feature Selection and Multinomial Logistic Regression for Classification of Bearing Faults Using Compressively Sampled Vibration Signals. Entropy
**2022**, 24, 511. [Google Scholar] [CrossRef] - Tabrizi, A.; Garibaldi, L.; Fasana, A.; Marchesiello, S. Early damage detection of roller bearings using wavelet packet decomposition, ensemble empirical mode decomposition and support vector machine. Meccanica
**2015**, 50, 865–874. [Google Scholar] [CrossRef] - Han, H.; Cho, S.; Kwon, S.; Cho, S.B. Fault diagnosis using improved complete ensemble empirical mode decomposition with adaptive noise and power-based intrinsic mode function selection algorithm. Electronics
**2018**, 7, 16. [Google Scholar] [CrossRef] - Ge, J.; Niu, T.; Xu, D.; Yin, G.; Wang, Y. A Rolling Bearing Fault Diagnosis Method Based on EEMD-WSST Signal Reconstruction and Multi-Scale Entropy. Entropy
**2020**, 22, 290. [Google Scholar] [CrossRef] - Kumar, P.S.; Kumaraswamidhas, L.A.; Laha, S.K. Selecting effective intrinsic mode functions of empirical mode decomposition and variational mode decomposition using dynamic time warping algorithm for rolling element bearing fault diagnosis. Trans. Inst. Meas. Control
**2019**, 41, 1923–1932. [Google Scholar] [CrossRef] - Tang, G.; Luo, G.; Zhang, W.; Yang, C.; Wang, H. Underdetermined blind source separation with variational mode decomposition for compound roller bearing fault signals. Sensors
**2016**, 16, 897. [Google Scholar] [CrossRef][Green Version] - Zhang, C.; Wang, Y.; Deng, W. Fault Diagnosis for Rolling Bearings Using Optimized Variational Mode Decomposition and Resonance Demodulation. Entropy
**2020**, 22, 739. [Google Scholar] [CrossRef] - Rilling, G.; Flandrin, P.; Goncalves, P. On empirical mode decomposition and its algorithms. In Proceedings of the IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing, Grado, Italy, 8–11 June 2003; Volume 3, pp. 8–11. [Google Scholar]
- Yu, D.; Cheng, J.; Yang, Y. Application of EMD method and Hilbert spectrum to the fault diagnosis of roller bearings. Mech. Syst. Signal Process.
**2005**, 19, 259–270. [Google Scholar] [CrossRef] - Moore, K.J.; Kurt, M.; Eriten, M.; McFarland, D.M.; Bergman, L.A.; Vakakis, A.F. Wavelet-bounded empirical mode decomposition for measured time series analysis. Mech. Syst. Signal Process.
**2018**, 99, 14–29. [Google Scholar] [CrossRef] - Jolliffe, I.T. Principal Component Analysis; Springer: New York, NY, USA, 2002. [Google Scholar]
- Shawe-Taylor, J.; Cristianini, N. Kernel Methods for Pattern Analysis; Cambridge University Press: Cambridge, UK, 2004. [Google Scholar]
- Mucherino, A.; Papajorgji, P.J.; Pardalos, P.M. k-Nearest Neighbor Classification. In Data Mining in Agriculture; Mucherino, A., Papajorgji, P.J., Pardalos, P.M., Eds.; Springer: New York, NY, USA, 2009; pp. 83–106. [Google Scholar]
- Izenman, A.J. Modern Multivariate Statistical Techniques: Regression, Classification, and Manifold Learning; Springer: New York, NY, USA, 2008. [Google Scholar]
- Webb, G.I. Naïve Bayes. In Encyclopedia of Machine Learning; Sammut, C., Webb, G.I., Eds.; Springer: Boston, MA, USA, 2010; pp. 713–714. [Google Scholar]
- Case Western Reserve University, USA. Bearing Data Centre. 2020. Available online: http://csegroups.case.edu/bearingdatacenter/pages/download-data-file/ (accessed on 2 March 2020).
- Harmouche, J.; Delpha, C.; Diallo, D. Incipient fault detection and diagnosis based on Kullback–Leibler divergence using principal component analysis: Part I. Signal Process.
**2014**, 94, 278–287. [Google Scholar] [CrossRef] - Cai, P.; Deng, X. Incipient fault detection for nonlinear processes based on dynamic multi-block probability related kernel principal component analysis. ISA Trans.
**2020**, 105, 210–220. [Google Scholar] [CrossRef] - Delpha, C.; Diallo, D. Kullback—Leibler divergence for incipient fault diagnosis. In Signal Processing for Fault Detection and Diagnosis in Electric Machines and Systems; Benbouzid, M., Ed.; IET, The Institution of Engineering and Technology: London, UK, 2020; pp. 87–118. [Google Scholar]
- Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. Bearing fault detection using intrinsic mode functions statistical information. In Proceedings of the 2018 15th International Multi-Conference on Systems, Signals & Devices (SSD), Yasmine Hammamet, Tunisia, 19–22 March 2018; pp. 870–875. [Google Scholar]
- Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. Intrinsic Mode Function Selection and Statistical Information Analysis for Bearing Ball Fault Detection. In Diagnosis, Fault Detection & Tolerant Control; Derbel, N., Ghommam, J., Zhu, Q., Eds.; Springer: Singapore, Singapore, 2020; pp. 111–135. [Google Scholar]
- Fawcett, T. An introduction to ROC analysis. Pattern Recognit. Lett.
**2006**, 27, 861–874. [Google Scholar] [CrossRef] - Schölkopf, B.; Smola, A.; Müller, K.R. Kernel principal component analysis. In Proceedings of the International Conference on Artificial Neural Networks; Springer: Berlin/Heidelberg, Germany, 1997; pp. 583–588. [Google Scholar]
- Lu, H.; Meng, Y.; Yan, K.; Gao, Z. Kernel principal component analysis combining rotation forest method for linearly inseparable data. Cogn. Syst. Res.
**2019**, 53, 111–122. [Google Scholar] [CrossRef] - Dong, S.; Luo, T.; Zhong, L.; Chen, L.; Xu, X. Fault diagnosis of bearing based on the kernel principal component analysis and optimized k-nearest neighbour model. J. Low Freq. Noise Vib. Act. Control
**2017**, 36, 354–365. [Google Scholar] [CrossRef] - Wu, G.; Yuan, H.; Gao, B.; Li, S. Fault diagnosis of power transformer based on feature evaluation and kernel principal component analysis. High Volt. Eng.
**2017**, 43, 2533–2540. [Google Scholar] - Zhang, X.; Delpha, C. Improved Incipient Fault Detection Using Jensen-Shannon Divergence and KPCA. In Proceedings of the 2020 Prognostics and Health Management Conference (PHM 2020), Besancon, France, 4–7 May 2020; pp. 241–246. [Google Scholar]
- Fauvel, M.; Chanussot, J.; Benediktsson, J.A. Kernel principal component analysis for the classification of hyperspectral remote sensing data over urban areas. EURASIP J. Adv. Signal Process.
**2009**, 2009, 783194. [Google Scholar] [CrossRef] - Cortes, C.; Vapnik, V. Support-vector networks. Mach. Learn.
**1995**, 20, 273–297. [Google Scholar] [CrossRef] - Wu, Y.; Ianakiev, K.; Govindaraju, V. Improved k-Nearest Neighbor classification. Pattern Recognit.
**2002**, 35, 2311–2318. [Google Scholar] [CrossRef] - Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. Bearing Fault Severity Classification Based on EMD-KLD: A Comparative Study for Incipient Ball Faults. In Proceedings of the 2020 Prognostics and Health Management Conference (PHM-Besançon), Besancon, France, 4–7 May 2020; pp. 257–262. [Google Scholar]
- Safavian, S.R.; Landgrebe, D. A survey of decision tree classifier methodology. IEEE Trans. Syst. Man Cybern.
**1991**, 21, 660–674. [Google Scholar] [CrossRef] - Polat, K.; Güneş, S. Classification of epileptiform EEG using a hybrid system based on decision tree classifier and fast Fourier transform. Appl. Math. Comput.
**2007**, 187, 1017–1026. [Google Scholar] [CrossRef] - Saimurugan, M.; Ramachandran, K.; Sugumaran, V.; Sakthivel, N. Multi component fault diagnosis of rotational mechanical system based on decision tree and support vector machine. Expert Syst. Appl.
**2011**, 38, 3819–3826. [Google Scholar] [CrossRef] - Zhang, N.; Wu, L.; Yang, J.; Guan, Y. Naive bayes bearing fault diagnosis based on enhanced independence of data. Sensors
**2018**, 18, 463. [Google Scholar] [CrossRef] - Addin, O.; Sapuan, S.; Othman, M.; Ali, B.A. Comparison of Naïve bayes classifier with back propagation neural network classifier based on f-folds feature extraction algorithm for ball bearing fault diagnostic system. Int. J. Phys. Sci.
**2011**, 6, 3181–3188. [Google Scholar] - Vernekar, K.; Kumar, H.; Gangadharan, K. Engine gearbox fault diagnosis using empirical mode decomposition method and Naïve Bayes algorithm. Sādhanā
**2017**, 42, 1143–1153. [Google Scholar] [CrossRef] - Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. A comparative study for ball bearing fault classification using kernel-SVM with Kullback Leibler divergence selected features. In Proceedings of the IECON 2019-45th Annual Conference of the IEEE Industrial Electronics Society, Lisbon, Portugal, 14–17 October 2019; Volume 1, pp. 6969–6974. [Google Scholar]
- Sharma, A.; Amarnath, M.; Kankar, P. Feature extraction and fault severity classification in ball bearings. J. Vib. Control
**2016**, 22, 176–192. [Google Scholar] [CrossRef] - Attoui, I.; Fergani, N.; Boutasseta, N.; Oudjani, B.; Deliou, A. A new time–frequency method for identification and classification of ball bearing faults. J. Sound Vib.
**2017**, 397, 241–265. [Google Scholar] [CrossRef] - Babouri, M.K.; Djebala, A.; Ouelaa, N.; Oudjani, B.; Younes, R. Rolling bearing faults severity classification using a combined approach based on multi-scales principal component analysis and fuzzy technique. Int. J. Adv. Manuf. Technol.
**2020**, 107, 4301–4316. [Google Scholar] [CrossRef]

**Figure 3.**Testbed of the CWRU for bearing defects [42] and the components of REBs: (

**a**) Photo of the test bench, (

**b**) Structural description of the bench.

**Figure 4.**The three-dimensional principal subspace for bearing ball fault data under the different load conditions.

**Figure 5.**KPCA kernel function hyperparameters adjustment under fully loaded condition. (

**a**) Gaussian kernel width parameter regularisation; (

**b**) polynomial kernel degree parameter regularisation.

**Figure 6.**KPCA scatter plot under the all-load-conditions combination. (

**a**) Results with Gaussian Kernel ($\gamma =0.01$); (

**b**) results with polynomial kernel ($p=6$).

**Figure 7.**Classification time computation for the $C4$ feature selection. (

**a**) Training time evaluation; (

**b**) testing time evaluation.

IMF Rank | SNR RDP (%) |
---|---|

1 | 5 |

2 | 11.7 |

3 | 21.6 |

4 | 20 |

5 | 17.3 |

6 | 22.3 |

7 | 28.2 |

8 | 34.8 |

9 | 44.9 |

10 | 51.8 |

11 | 59.2 |

12 | 62.1 |

13 | 68.2 |

14 | 68.9 |

15 | 59.1 |

16 | 62.1 |

17 | 83 |

18 | 81.1 |

AUC for Mean | AUC for Variance | AUC for Skewness | AUC for Kurtosis | AUC for KLD | ||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

IMF | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ |

2 | $0.971$ | $0.982$ | 1 | $0.998$ | 1 | $0.996$ | 1 | 1 | $0.7$ | $0.838$ | $0.835$ | $0.65$ | 1 | $0.955$ | $0.996$ | $0.878$ | 1 | 1 | 1 | 1 |

3 | $0.853$ | $0.968$ | $0.602$ | $0.836$ | 1 | $0.998$ | 1 | 1 | $0.816$ | $0.996$ | $0.911$ | $0.962$ | $0.824$ | $0.845$ | $0.784$ | $0.638$ | 1 | 1 | 1 | 1 |

4 | $0.937$ | $0.82$ | $0.801$ | $0.84$ | 1 | $0.968$ | 1 | 1 | $0.882$ | $0.733$ | $0.872$ | $0.602$ | 1 | $0.975$ | 1 | $0.993$ | 1 | 1 | 1 | 1 |

5 | $0.586$ | $0.773$ | $0.999$ | $0.729$ | 1 | $0.967$ | 1 | 1 | $0.587$ | $0.554$ | $0.674$ | $0.963$ | $0.926$ | $0.966$ | 1 | 1 | 1 | 1 | 1 | 1 |

6 | $0.889$ | $0.71$ | $0.979$ | $0.953$ | 1 | $0.995$ | 1 | 1 | $0.776$ | $0.681$ | $0.634$ | $0.705$ | 1 | $0.769$ | $0.917$ | $0.913$ | 1 | 1 | 1 | 1 |

7 | $0.649$ | $0.634$ | $0.783$ | $0.958$ | 1 | $0.962$ | 1 | 1 | $0.565$ | $0.51$ | $0.632$ | $0.6$ | $0.968$ | $0.862$ | $0.994$ | $0.733$ | $0.666$ | $0.457$ | $0.842$ | $0.911$ |

8 | $0.517$ | $0.623$ | $0.637$ | $0.651$ | 1 | $0.972$ | 1 | 1 | $0.552$ | $0.687$ | $0.674$ | $0.622$ | $0.534$ | $0.733$ | $0.917$ | $0.976$ | $0.49$ | $0.45$ | $0.901$ | $0.5$ |

_{0}: no-load condition, L

_{1}: half-load condition, L

_{2}: full-load condition, L

_{3}: overload condition.

Load Condition | PC | Eigenvalue | Variance Contribution (%) | Cumulative Variance (%) |
---|---|---|---|---|

${L}_{0}$ | 1 | 2.638 | 65.956 | 65.956 |

2 | 0.982 | 24.572 | 90.53 | |

3 | 0.277 | 6.932 | 97.46 | |

4 | 0.105 | 2.537 | 100 | |

${L}_{1}$ | 1 | 1.672 | 41.814 | 41.814 |

2 | 1.475 | 36.847 | 78.69 | |

3 | 0.482 | 12.071 | 90.76 | |

4 | 0.369 | 9.239 | 100 | |

${L}_{2}$ | 1 | 3.073 | 75.927 | 75.927 |

2 | 0.744 | 18.617 | 94.55 | |

3 | 0.136 | 3.422 | 97.97 | |

4 | 0.081 | 2.031 | 100 | |

${L}_{3}$ | 1 | 2.706 | 67.653 | 67.653 |

2 | 1.039 | 25.996 | 93.65 | |

3 | 0.172 | 4.31 | 97.96 | |

4 | 0.081 | 2.01 | 100 | |

${L}_{n}$ | 1 | 1.829 | 45.747 | 45.747 |

2 | 1.132 | 28.322 | 74.07 | |

3 | 0.806 | 20.151 | 94.22 | |

4 | 0.231 | 5.779 | 100 |

KPC | Variance Contribution | Cumulative Variance (%) | ||
---|---|---|---|---|

Gaussian Kernel | Polynomial Kernel | Gaussian Kernel | Polynomial Kernel | |

1 | $0.433$ | $0.967$ | $\mathbf{43.3}$ | $\mathbf{96.7}$ |

2 | $0.281$ | $0.23$ | $\mathbf{71.4}$ | $\mathbf{99}$ |

3 | $0.199$ | $0.007$ | $\mathbf{91.3}$ | $99.7$ |

4 | $0.087$ | $0.003$ | 100 | 100 |

Classifier | KNN | DT | NB | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

Load (hp) | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ |

Training accuracy rate (%) | $98.2$ | $99.8$ | $99.1$ | 100 | 99 | $98.3$ | 100 | $98.6$ | 100 | $98.6$ | 98 | $99.3$ | 98 | $99.6$ | 82.3 |

Testing accuracy rate (%) | $97.42$ | $99.92$ | $98.91$ | $99.75$ | $98.83$ | $98.25$ | $99.92$ | $98.33$ | $99.75$ | $98.15$ | $\mathbf{96.5}$ | $99.42$ | $97.83$ | $99.83$ | 81.92 |

Training time (s) | $0.31$ | $0.9$ | $0.31$ | $0.3$ | $0.56$ | $0.28$ | $0.24$ | $0.25$ | $0.25$ | $1.15$ | $0.65$ | $0.75$ | $0.7$ | $0.67$ | 1 |

Testing time (s) | $0.03$ | $0.02$ | $0.03$ | $0.02$ | $0.15$ | $0.02$ | $0.02$ | $0.01$ | $0.02$ | $0.05$ | $0.05$ | $0.02$ | $0.02$ | $0.02$ | $0.05$ |

Features | Relevant $\mathbf{IMFs}$ | |||
---|---|---|---|---|

Variance | $IM{F}_{2}$ | $IM{F}_{3}$ | $IM{F}_{4}$ | |

KLD | $IM{F}_{2}$ | $IM{F}_{3}$ | $IM{F}_{4}$ | $IM{F}_{6}$ |

Classifier | KNN | DT | NB | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

Load (hp) | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ |

Training accuracy rate (%) | 100 | 100 | $99.9$ | 100 | 100 | 100 | 100 | $99.9$ | 100 | 100 | $99.9$ | 100 | $99.9$ | 100 | 100 |

Testing accuracy rate (%) | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | $99.83$ | 100 | $99.85$ | 100 | 100 |

Training time (s) | $0.89$ | $1.1$ | $1.31$ | $0.88$ | $0.69$ | $1.58$ | $1.04$ | $0.55$ | $0.56$ | $0.55$ | $1.8$ | $0.87$ | $1.24$ | $1.01$ | $1.53$ |

Testing time (s) | $0.19$ | $0.05$ | $0.03$ | $0.04$ | $0.06$ | $0.05$ | $0.03$ | $0.03$ | $0.02$ | $0.04$ | $0.07$ | $0.04$ | $0.02$ | $0.02$ | $0.05$ |

Ref | Fault Type | Ball | ||||||
---|---|---|---|---|---|---|---|---|

Load (hp) | ${\mathit{L}}_{0}$ | ${\mathit{L}}_{1}$ | ${\mathit{L}}_{2}$ | ${\mathit{L}}_{3}$ | ${\mathit{L}}_{\mathit{n}}$ | Mean | ||

Algorithm | Testing Accuracy Rates (%) | |||||||

[6] | MPE | KNN | 93 | 99 | 100 | 100 | Not provided | 98 |

SVM | 81 | 99 | 100 | 98 | Not provided | 94.5 | ||

Logic regression | 96 | 99 | 100 | 100 | Not provided | 98.75 | ||

Backpropagation NN | 70 | 91 | 90 | 93 | Not provided | 86 | ||

Extreme learning Machine | 92 | 90 | 100 | 100 | Not provided | 97.5 | ||

Soft regression | 94 | 99 | 100 | 100 | Not provided | 98.25 | ||

Proposed technique | KLD and variance | KNN | 100 | 100 | 100 | 100 | 100 | 100 |

DT | 100 | 100 | 100 | 100 | 100 | 100 | ||

NB | 99.83 | 100 | 99.85 | 100 | 100 | 99.92 |

**Table 9.**Best feature combinations according to the classification accuracy under different load conditions.

KNN | DT | NB | ||||
---|---|---|---|---|---|---|

Load Condition | $\mathit{TrA}\phantom{\rule{0.166667em}{0ex}}(\%)$ | $\mathit{TsA}\phantom{\rule{0.166667em}{0ex}}(\%)$ | $\mathit{TrA}\phantom{\rule{0.166667em}{0ex}}(\%)$ | $\mathit{TsA}\phantom{\rule{0.166667em}{0ex}}(\%)$ | $\mathit{TrA}\phantom{\rule{0.166667em}{0ex}}(\%)$ | $\mathit{TsA}\phantom{\rule{0.166667em}{0ex}}(\%)$ |

${L}_{0}$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ |

$C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | |

$C{2}_{3}$ | $C{2}_{3}$ | |||||

${L}_{1}$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ |

$C{2}_{3}$ | $C{2}_{2}$ | $C{2}_{2}$ | $C{2}_{2}$ | $C{2}_{2}$ | $C{2}_{2}$ | |

$C{1}_{2}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | |

$C{2}_{4}$ | $C{1}_{2}$ | $C{1}_{2}$ | ||||

$C{1}_{2}$ | ||||||

${L}_{2}$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ |

$C{2}_{3}$ | $C{2}_{2}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | |

$C{1}_{2}$ | $C{2}_{3}$ | $C{1}_{2}$ | $C{1}_{2}$ | $C{1}_{2}$ | ||

$C{1}_{1}$ | ||||||

$C{1}_{2}$ | ||||||

${L}_{3}$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ |

$C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | $C{2}_{1}$ | |

$C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | $C{2}_{3}$ | |

$C{1}_{1}$ | $C{2}_{4}$ | $C{1}_{1}$ | $C{1}_{1}$ | $C{1}_{1}$ | $C{1}_{1}$ | |

$C{1}_{1}$ | ||||||

${L}_{n}$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ | $C4$ |

$C{2}_{1}$ | $C{2}_{1}$ | |||||

$C{2}_{3}$ | $C{2}_{2}$ | |||||

$C{2}_{3}$ | ||||||

$C{2}_{4}$ | ||||||

$C{1}_{1}$ | ||||||

$C{1}_{2}$ | ||||||

$C{1}_{4}$ |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Mezni, Z.; Delpha, C.; Diallo, D.; Braham, A. Performance of Bearing Ball Defect Classification Based on the Fusion of Selected Statistical Features. *Entropy* **2022**, *24*, 1251.
https://doi.org/10.3390/e24091251

**AMA Style**

Mezni Z, Delpha C, Diallo D, Braham A. Performance of Bearing Ball Defect Classification Based on the Fusion of Selected Statistical Features. *Entropy*. 2022; 24(9):1251.
https://doi.org/10.3390/e24091251

**Chicago/Turabian Style**

Mezni, Zahra, Claude Delpha, Demba Diallo, and Ahmed Braham. 2022. "Performance of Bearing Ball Defect Classification Based on the Fusion of Selected Statistical Features" *Entropy* 24, no. 9: 1251.
https://doi.org/10.3390/e24091251