# Color Recurrence Plots for Bearing Fault Diagnosis

^{1}

^{2}

^{3}

^{4}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Preliminaries

#### 2.1. Non-Uniform Embedding

#### 2.2. The Standard Recurrence Plot

#### 2.3. Recurrence Plots for Bearing Fault Diagnosis

## 3. The Color Recurrence Plot

#### 3.1. The Proposed Recurrence Plot Based on a Fixed Time Lag

#### 3.2. The Color Recurrence Plot Based on Non-Uniform Embedding

#### 3.3. Color Recurrence Plots Produced by Bearing Vibration Signals

## 4. The Description of the Data Set

## 5. Transfer Learning for Bearing Fault Diagnosis

#### 5.1. Transfer Learning Definition

_{++}or Java and expose him to programming in python. The person would be able to use the skills acquired in learning C

_{++}or Java and would be able to learn python much faster compared to a person who has no previous programming knowledge.

#### 5.2. Pretrained Network Selection

#### 5.3. Transfer Learning for AlexNet and SqueezeNet

#### 5.3.1. Transfer Learning for AlexNet

#### 5.3.2. Transfer Learning for SqueezeNet

`‘classificationLayer_preidction’`contain information on how to combine the features that the network uses for class probabilities. For retraining of the SqueezeNet the last two layers are replaced with a new set of layers to adapt to the new data set. In SqueezeNet, the last learnable layer is the final convolutional layer instead, which needs to be replaced by a new convolutional layer with a number of filters equal to 2 for the classification of the new data set with crack and without crack. Some learning parameters are then adjusted to, e.g., Weightlearnfactor and Biaslearnfactor. The initial learning rate parameter is also adjusted to slow down the learning rate of the new network to increase its prediction capabilities. Now, the new network is retrained with a new data set of images.

#### 5.3.3. Network Parameter Optimization

- Identify the most impacting network parameter through a sensitivity analysis.
- For sensitivity analysis, a parameter range is decided and then prediction is performed using the set of network parameter impact on prediction accuracy is checked.
- Several combinations of different network parameters are tested.
- Finally, the network parameters giving the best prediction accuracy are selected

## 6. Color Recurrence Plot Analysis Using Transfer Learning

## 7. Conclusions and Discussions

## Author Contributions

## Funding

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Kankar, P.; Sharma, S.C.; Harsha, S. Fault diagnosis of ball bearings using machine learning methods. Expert Syst. Appl.
**2011**, 38, 1876–1886. [Google Scholar] [CrossRef] - Azadeh, A.; Saberi, M.; Kazem, A.; Ebrahimipour, V.; Nourmohammadzadeh, A.; Saberi, Z. A flexible algorithm for fault diagnosis in a centrifugal pump with corrupted data and noise based on ANN and support vector machine with hyper-parameters optimization. Appl. Soft Comput.
**2013**, 13, 1478–1485. [Google Scholar] [CrossRef] - Hamadache, M.; Jung, J.H.; Park, J.; Youn, B.D. A comprehensive review of artificial intelligence-based approaches for rolling element bearing PHM: Shallow and deep learning. JMST Adv.
**2019**, 1, 125–151. [Google Scholar] [CrossRef] [Green Version] - Liu, R.; Yang, B.; Zio, E.; Chen, X. Artificial intelligence for fault diagnosis of rotating machinery: A review. Mech. Syst. Signal Process.
**2018**, 108, 33–47. [Google Scholar] [CrossRef] - Hoang, D.T.; Kang, H.J. A survey on Deep Learning based bearing fault diagnosis. Neurocomputing
**2019**, 335, 327–335. [Google Scholar] [CrossRef] - Eren, L.; Ince, T.; Kiranyaz, S. A Generic Intelligent Bearing Fault Diagnosis System Using Compact Adaptive 1D CNN Classifier. J. Signal Process. Syst.
**2019**, 91, 179–189. [Google Scholar] [CrossRef] - He, M.; He, D. A new hybrid deep signal processing approach for bearing fault diagnosis using vibration signals. Neurocomputing
**2020**, 396, 542–555. [Google Scholar] [CrossRef] - Xu, G.; Liu, M.; Jiang, Z.; Söffker, D.; Shen, W. Bearing Fault Diagnosis Method Based on Deep Convolutional Neural Network and Random Forest Ensemble Learning. Sensors
**2019**, 19, 1088. [Google Scholar] [CrossRef] [Green Version] - Leite, G.; Araújo, A.M.; Rosas, P.A.C.; Stosic, T.; Stosic, B. Entropy measures for early detection of bearing faults. Phys. A Stat. Mech. Appl.
**2019**, 514, 458–472. [Google Scholar] [CrossRef] - Li, Y.; Xu, M.; Wei, Y.; Huang, W. A new rolling bearing fault diagnosis method based on multiscale permutation entropy and improved support vector machine based binary tree. Measurement
**2016**, 77, 80–94. [Google Scholar] [CrossRef] - Jiang, W.; Li, Z.; Jiang, A.; Lei, Y.; Wang, H. Recurrence Plot Quantitative Analysis-Based Fault Recognition Method of Rolling Bearing. In Proceedings of the 2019 Prognostics and System Health Management Conference (PHM-Qingdao), Qingdao, China, 25–27 October 2019; pp. 1–8. [Google Scholar] [CrossRef]
- Zhou, C.; Zhang, W. Recurrence Plot Based Damage Detection Method by Integrating Control Chart. Entropy
**2015**, 17, 2624–2641. [Google Scholar] [CrossRef] [Green Version] - Rajesh, V.; Rajesh, M. Bearing fault feature extraction by recurrence quantification analysis. World Acad. Sci. Eng. Technol.
**2008**, 45, 449–453. [Google Scholar] - Sharma, A. Fault Diagnosis of Bearings Using Recurrences and Artificial Intelligence Techniques. ASME J. Nondestruct. Eval.
**2022**, 5, 031004. [Google Scholar] [CrossRef] - Chen, J.; Wang, J.; Zhu, J.; Lee, T.H.; de Silva, C.W. Unsupervised Cross-Domain Fault Diagnosis Using Feature Representation Alignment Networks for Rotating Machinery. IEEE/ASME Trans. Mechatron.
**2021**, 26, 2770–2781. [Google Scholar] [CrossRef] - Zhang, Y.; Ren, Z.; Zhou, S.; Feng, K.; Yu, K.; Liu, Z. Supervised Contrastive Learning-Based Domain Adaptation Network for Intelligent Unsupervised Fault Diagnosis of Rolling Bearing. IEEE/ASME Trans. Mechatron.
**2022**, 1–10. [Google Scholar] [CrossRef] - Zhang, S.; Zhang, S.; Wang, B.; Habetler, T.G. Deep Learning Algorithms for Bearing Fault Diagnostics—A Comprehensive Review. arXiv
**2020**, arXiv:1901.08247v3. [Google Scholar] [CrossRef] - Sauer, T.; Yorke, J.A.; Casdagli, M. Embedology. J. Stat. Phys.
**1991**, 65, 579–616. [Google Scholar] [CrossRef] - Schreiber, T. Interdisciplinary application of nonlinear time series methods. Phys. Rep.
**1999**, 308, 1–64. [Google Scholar] [CrossRef] [Green Version] - Cao, L. Practical method for determining the minimum embedding dimension of a scalar time series. Phys. D Nonlinear Phenom.
**1997**, 110, 43–50. [Google Scholar] [CrossRef] - Buzug, T.; Pfister, G. Optimal delay time and embedding dimension for delay-time coordinates by analysis of the global static and local dynamical behavior of strange attractors. Phys. Rev. A
**1992**, 45, 7073–7084. [Google Scholar] [CrossRef] - Manabe, Y.; Chakraborty, B. A novel approach for estimation of optimal embedding parameters of nonlinear time series by structural learning of neural network. Neurocomputing
**2007**, 70, 1360–1371. [Google Scholar] [CrossRef] - Ragulskis, M.; Lukoseviciute, K. Non-uniform attractor embedding for time series forecasting by fuzzy inference systems. Neurocomputing
**2009**, 72, 2618–2626. [Google Scholar] [CrossRef] - Jia, Z.; Lin, Y.; Liu, Y.; Jiao, Z.; Wang, J. Refined nonuniform embedding for coupling detection in multivariate time series. Phys. Rev. E
**2020**, 101, 062113. [Google Scholar] [CrossRef] [PubMed] - Zhang, J. Low-dimensional approximation searching strategy for transfer entropy from non-uniform embedding. PLoS ONE
**2018**, 13, e0194382. [Google Scholar] [CrossRef] [Green Version] - Lukoseviciute, K.; Ragulskis, M. Evolutionary algorithms for the selection of time lags for time series forecasting by fuzzy inference systems. Neurocomputing
**2010**, 73, 2077–2088. [Google Scholar] [CrossRef] - Kugiumtzis, D. Direct-coupling information measure from nonuniform embedding. Phys. Rev. E
**2013**, 87, 062918. [Google Scholar] [CrossRef] [Green Version] - Faes, L.; Nollo, G.; Porta, A. Information-based detection of nonlinear Granger causality in multivariate processes via a nonuniform embedding technique. Phys. Rev. E
**2011**, 83, 051112. [Google Scholar] [CrossRef] [Green Version] - Gómez-García, J.A.; Godino-Llorente, J.I.; Castellanos-Dominguez, G. Non uniform Embedding based on Relevance Analysis with reduced computational complexity: Application to the detection of pathologies from biosignal recordings. Neurocomputing
**2014**, 132, 148–158. [Google Scholar] [CrossRef] [Green Version] - Timofejeva, I.; Poskuviene, K.; Cao, M.; Ragulskis, M. Synchronization Measure Based on a Geometric Approach to Attractor Embedding Using Finite Observation Windows. Complexity
**2018**, 2018, 8259496. [Google Scholar] [CrossRef] - Eckmann, J.P.; Kamphorst, S.O.; Ruelle, D. Recurrence Plots of Dynamical Systems. Europhys. Lett.
**1987**, 4, 973–977. [Google Scholar] [CrossRef] - Marwan, N.; Carmen Romano, M.; Thiel, M.; Kurths, J. Recurrence plots for the analysis of complex systems. Phys. Rep.
**2007**, 438, 237–329. [Google Scholar] [CrossRef] - Case Western Reserve University Bearind Data Center. 2022. Available online: https://engineering.case.edu/bearingdatacenter/download-data-file (accessed on 2 August 2022).
- Kennel, M.B.; Brown, R.; Abarbanel, H.D.I. Determining embedding dimension for phase-space reconstruction using a geometrical construction. Phys. Rev. A
**1992**, 45, 3403–3411. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Weiss, K.; Khoshgoftaar, T.M.; Wang, D. A survey of transfer learning. J. Big Data
**2016**, 3, 9. [Google Scholar] [CrossRef] [Green Version] - Iandola, F.N.; Moskewicz, M.W.; Ashraf, K.; Han, S.; Dally, W.J.; Keutzer, K. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size. arXiv
**2016**, arXiv:1602.07360. [Google Scholar] - Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet Classification with Deep Convolutional Neural Networks. Commun. ACM
**2017**, 60, 84–90. [Google Scholar] [CrossRef] [Green Version] - Pretrained Deep Neural Networks. 30 August 2021. Available online: https://ch.mathworks.com/help/deeplearning/ug/pretrained-convolutional-neural-networks.html (accessed on 2 August 2022).
- Szegedy, C.; Ioffe, S.; Vanhoucke, V.; Alemi, A.A. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning. In Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence, AAAI’17, San Francisco, CA, USA, 4–9 February 2017; pp. 4278–4284. [Google Scholar]
- Palevičius, P.; Pal, M.; Landauskas, M.; Orinaitė, U.; Timofejeva, I.; Ragulskis, M. Automatic Detection of Cracks on Concrete Surfaces in the Presence of Shadows. Sensors
**2022**, 22, 3662. [Google Scholar] [CrossRef] - Pal, M.; Palevičius, P.; Landauskas, M.; Orinaitė, U.; Timofejeva, I.; Ragulskis, M. An Overview of Challenges Associated with Automatic Detection of Concrete Cracks in the Presence of Shadows. Appl. Sci.
**2021**, 11, 1396. [Google Scholar] [CrossRef]

**Figure 1.**The standard recurrence plot generated by a normal bearing vibration signal without a fault (the dataset of Case Western Reserve University School of Engineering [33]). The threshold parameter $\epsilon $ is set to 0.253 which results in an equal proportion of black and white pixels.

**Figure 2.**The standard recurrence plot of a normal bearing vibration signal without a fault. The line graph shows the relationship between the percentage of black pixels in the recurrence plot from the threshold parameter $\epsilon $. The ratio between black and white pixels is equal at $\epsilon =0.253$. Part (

**A**) shows the standard recurrence plot at $\epsilon =0.21$; part (

**B**) at $\epsilon =0.144$; part (

**C**) at $\epsilon =0.295$.

**Figure 3.**Figure showing 3 different recurrence plot generation algorithms. Standard recurrence plot, Algorithm A corresponds to a recurrence relationship involving non-uniform time delay with time delay $\tau $ $+/-$ 1, Algorithm B corresponds to the recurrence relationship involving purely non-uniform time delays $\tau $.

**Figure 4.**A schematic diagram illustrating non-uniform embedding of a scalar time series into a 5-dimensional delay-coordinate space. ${X}_{1},\dots ,{X}_{5}$ denote the axes of the reconstructed phase space; ${\tau}_{1},\dots ,{\tau}_{4}$ denote time delays.

**Figure 5.**A schematic diagram illustrating the formation of the Color recurrence plot for a normal bearing test data without a fault. The embedding dimension $d=5$ yields ten different dichotomous recurrence plots based on a fixed time lag. The first column represents recurrence plots with a single time delay (${\tau}_{1}$, ${\tau}_{2}$, ${\tau}_{3}$, and ${\tau}_{4}$). The second column represents recurrence plots with double time delays (${\tau}_{1}+{\tau}_{2}$, ${\tau}_{2}+{\tau}_{3}$, and ${\tau}_{3}+{\tau}_{4}$). The third column represents triple time delays (${\tau}_{1}+{\tau}_{2}+{\tau}_{3}$, and ${\tau}_{2}+{\tau}_{3}+{\tau}_{4}$). The fourth column represents a recurrence plot with the maximal time delay (${\tau}_{1}+{\tau}_{2}+{\tau}_{3}+{\tau}_{4}$). The Color recurrence plot is produced by the arithmetic averaging of dichotomous plots; the number of different colors is 11. The Color recurrence plot is enlarged for clarity.

**Figure 6.**Color recurrence plots generated by vibration signals of a ball bearing. Part (

**A**) shows the Color recurrence plot produced by a normal bearing without a fault. Parts (

**B**–

**D**) show recurrent plots generated by the bearing with the ball fault, the inner race fault, and the outer race fault accordingly.

**Figure 10.**Figure showing application of network optimization workflow on AlexNet and SqueezeNet using the database of images and trial runs for initial learning rateparameter optimization. The blue box highlights the best case with optimum parameters selected from the experimental design.

**Figure 11.**Figure showing application of network optimization workflow on AlexNet and SqueezeNet using the database of images and trial runs with Initial Learning rate and Dropout Probability variations.

**Figure 13.**Figure showing conceptual workflow of how Alextnet or Squeezenet will be used for image classification using Color recurrence plots.

**Figure 14.**Figure showing training of SqueezeNet using Transfer learning on the Color recurrence plots combining the classes into a single classification problem.

**Figure 15.**Figure showing training of AlexNet using Transfer learning on the Color recurrence plots combining the classes into a single classification problem.

**Figure 16.**Figure showing Confusion Matrix for AlexNet and SqueezeNet using Transfer learning on the Color recurrence plots for the drive end bearing subset of the data set.

**Figure 17.**Figure showing Confusion Matrix for AlexNet and SqueezeNet using Transfer learning on the Color recurrence plots for the data set corresponding to the fan end bearing subset of the data set.

**Figure 18.**Figure showing Confusion Matrix for AlexNet and SqueezeNet using Transfer learning on the Color recurrence plots combining the classes into a single classification problem.

**Figure 19.**Figure showing confusion matrix using AlexNet and SqueezeNet for the fan end bearing data.

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**

Petrauskiene, V.; Pal, M.; Cao, M.; Wang, J.; Ragulskis, M.
Color Recurrence Plots for Bearing Fault Diagnosis. *Sensors* **2022**, *22*, 8870.
https://doi.org/10.3390/s22228870

**AMA Style**

Petrauskiene V, Pal M, Cao M, Wang J, Ragulskis M.
Color Recurrence Plots for Bearing Fault Diagnosis. *Sensors*. 2022; 22(22):8870.
https://doi.org/10.3390/s22228870

**Chicago/Turabian Style**

Petrauskiene, Vilma, Mayur Pal, Maosen Cao, Jie Wang, and Minvydas Ragulskis.
2022. "Color Recurrence Plots for Bearing Fault Diagnosis" *Sensors* 22, no. 22: 8870.
https://doi.org/10.3390/s22228870