Research on Deep Learning Method and Optimization of Vibration Characteristics of Rotating Equipment
<p>Technical roadmap of the article.</p> "> Figure 2
<p>NCEPU (Baoding) gear power transmission failure simulation bench.</p> "> Figure 3
<p>Faulty gear: (<b>a</b>) broken gear tooth; (<b>b</b>) gear crack; (<b>c</b>) gear wear and tear; (<b>d</b>) gear tooth deficiency.</p> "> Figure 4
<p>The CNN structure diagram.</p> "> Figure 5
<p>Feature information extracted by CNN in the previous layer of the fully connected layer (different sampling frequencies).</p> "> Figure 6
<p>t-SNE image of signal classification in the fully connected layer: (<b>a</b>) 2000 Hz, (<b>b</b>) 4000 Hz, (<b>c</b>) 8000 Hz, (<b>d</b>) 16,000 Hz, and (<b>e</b>) 32,000 Hz.</p> "> Figure 7
<p>Feature information extracted by CNN in the previous layer of the fully connected layer (different rotation speed).</p> "> Figure 8
<p>t-SNE image of signal classification in the fully connected layer: (<b>a</b>) 1200 rpm, (<b>b</b>) 1500 rpm, (<b>c</b>) 1800 rpm, and (<b>d</b>) 2100 rpm.</p> "> Figure 9
<p>Vibration signals at different sampling frequencies.</p> "> Figure 10
<p>Vibration signals at different rotation speeds.</p> "> Figure 11
<p>t-SNE image of signal classification in the fully connected layer: (<b>a</b>) 2000 Hz, (<b>b</b>) 4000 Hz, (<b>c</b>) 8000 Hz, (<b>d</b>) 16,000 Hz, and (<b>e</b>) 32,000 Hz.</p> "> Figure 12
<p>t-SNE image of signal classification in the fully connected layer: (<b>a</b>) 1200 rpm–2000 Hz, (<b>b</b>) 1500 rpm–2500 Hz, (<b>c</b>) 1800 rpm–3000 Hz, and (<b>d</b>) 2100 rpm–3500 Hz.</p> ">
Abstract
:1. Introduction
2. Introduction to CNN
2.1. Convolution Layer
2.2. Pool Layer
2.3. Full Connection Layer
3. Vibration State Recognition of Transmission System Based on CNN
3.1. Vibration Signal Sample Data Set
3.2. State Recognition Model Based on CNN and the Optimization of Convolution Kernels
4. The Matching of Convolution Kernel Scale and Signal
4.1. Feature Extraction of Signals at Different Sampling Frequencies
4.2. Feature Learning of Different Rotation Speed Signals
5. Model and Signal Matching Optimization Based on Optimal Perception Field
5.1. The Study of Convolution Kernel Optimization Based on Optimal Perception Field
5.2. Experimental Study
6. Experimental Comparison
7. Conclusions
- (1)
- The vibration signals of parallel gears with the same sampling frequency and rotation speed in the power engineering laboratory of NCEPU (Baoding) were taken as input and CNNs with different convolution kernel sizes were selected for fault recognition. Experimental results show that the fault identification accuracy of CNN for vibration signals is the highest (96.3%) when selecting Convalution1, Convalution2 as the convolution kernel of 1 × 8 and Convalution4, Convalution5 of 1 × 6. This also proves that there is a matching relation between the size of the convolution kernel and the recognition rate of the vibration signal.
- (2)
- Experiments using the CNN model show that when the optimal convolution kernel remains unchanged, the speed of vibration signal decreases, or the sampling frequency increases, the fault recognition rate of CNN decreases to varying degrees. Through the perceptual field optimization experiment, the relationship between the vibration signal speed and sampling frequency and the convolution kernel scale is obtained, and then the fault recognition rate of each vibration signal can be significantly improved. The results show that the accuracy of fault identification can be effectively improved by adjusting the matching between vibration signal and convolution kernel scale.
- (3)
- The optimal convolution kernel scale formula was derived from experiments in this paper. Experimental results showed that the optimal convolution kernel scale formula can improve the matching between the CNN model and the vibration signal, reduce the complexity of the CNN model, accelerate the calculation of CNN, as well as improve the recognition rate of CNN effectively. Finally, the fault recognition rate can reach 98% after optimization. At the same time, other methods in the literature were selected for comparative experiments to further prove the effectiveness of the convolution kernel optimization formula.
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Kong, Y.; Wang, T.; Chu, F. Meshing frequency modulation assisted empirical wavelet transform for fault diagnosis of wind turbine planetary ring gear. Renew. Energy 2019, 132, 1373–1388. [Google Scholar] [CrossRef]
- Jaouher, B.A.; Nader, F.; Lotfi, S.; Brigitte, C.; Farhat, F. Application of empirical mode decomposition and artificial neural network for automatic bearing fault diagnosis based on vibration signals. Appl. Acoust. 2015, 89, 16–27. [Google Scholar]
- Zhu, X.; Luo, X.; Zhao, J.; Hou, D.; Han, Z.; Wang, Y. Research on Deep Feature Learning and Condition Recognition Method for Bearing Vibration. Appl. Acoust. 2020, 168, 107435. [Google Scholar] [CrossRef]
- Yu, W.; Wang, Y.; Gui, W.; Yang, C. A Vibration Fault Diagnosis System of HGS Based on FNN. IFAC Proc. 2007, 39, 252–257. [Google Scholar] [CrossRef]
- Schmidhuber, J. Deep Learning in Neural Networks: An Overview. Neural Netw. 2015, 61, 85–117. [Google Scholar] [CrossRef] [Green Version]
- He, M.; He, D. Deep Learning Based Approach for Bearing Fault Diagnosis. IEEE Trans. Ind. Appl. 2017, 53, 3057–3065. [Google Scholar] [CrossRef]
- Li, D.; Yu, D. Deep learning: Methods and applications. Found. Trends Signal Process. 2014, 7, 197–387. [Google Scholar]
- Zhu, X.; Yuan, Y.; Zhou, P.; Xu, B.; Han, Z. An Improved Hilbert Vibration Decomposition Method for Analysis of Rotor Fault Signals. J. Braz. Soc. Mech. Sci. Eng. 2017, 39, 4921–4927. [Google Scholar] [CrossRef]
- Wang, L.; Liu, Z.; Miao, Q.; Zhang, X. Complete ensemble local mean decomposition with adaptive noise and its application to fault diagnosis for rolling bearings. Mech. Syst. Signal Process. 2018, 106, 24–39. [Google Scholar] [CrossRef]
- Pan, M.C.; Tsao, W.C. Using appropriate IMFs for envelope analysis in multiple fault diagnosis of ball bearings. Int. J. Mech. Sci. 2013, 69, 114–124. [Google Scholar] [CrossRef]
- Wu, Z.; Huang, N.E. Ensemble empirical mode decomposition: A noise assisted data analysis method. Adv. Adapt. Data Anal. 2011, 1, 0900004. [Google Scholar] [CrossRef]
- Smith, J.S. The local mean decomposition and its application to EEG perception data. J. R. Soc. Interface 2005, 2, 443–454. [Google Scholar] [CrossRef] [PubMed]
- Zhu, L.; Chen, Y.; Ghamisi, P.; Benediktsson, J.A. Generative Adversarial Networks for Hyperspectral Image Classification. IEEE Trans. Geosci. Remote Sens. 2018, 56, 5046–5063. [Google Scholar] [CrossRef]
- Wang, P.; Yan, R.; Gao, R.X. Virtualization and deep recognition for system fault classification. J. Manuf. Syst. 2017, 44, 310–316. [Google Scholar] [CrossRef]
- Li, Z.; Li, W.; Zhao, X. Feature Frequency Extraction Based on Principal Component Analysis and Its Application in Axis Orbit. Shock. Vib. 2018, 2018, 2530248. [Google Scholar] [CrossRef] [Green Version]
- Zhu, X.; Zhao, J.; Hou, D.; Han, Z. An SDP Characteristic Information Fusion-Based CNN Vibration Fault Diagnosis Method. Shock. Vib. 2019, 2019, 3926963. [Google Scholar] [CrossRef]
- Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet classification with deep convolutional neural networks. Int. Conf. Neural Inf. Process. Syst. 2012, 25, 1106–1114. [Google Scholar] [CrossRef]
- Simonyan, K.; Zisserman, A. Very Deeolutional Networks for Large-Scale Image Recognition. arXiv 2014, arXiv:1409.1556. [Google Scholar]
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 26 June–1 July 2016; pp. 770–778. [Google Scholar]
- Szegedy, C.; Ioffe, S.; Vanhoucke, V.; Alemi, A. Inception-v4, Inception-Resnet and the Impact of Residual Connections on Learning. arXiv 2016, arXiv:1602.07261. [Google Scholar]
- Sun, W.; Zhao, R.; Yan, R.; Shao, S.; Chen, X. Convolutional Discriminative Feature Learning for Induction Motor Fault Diagnosis. IEEE Trans. Ind. Inform. 2017, 13, 1350–1359. [Google Scholar] [CrossRef]
- Zhang, W.; Li, C.; Peng, G.; Chen, Y.; Zhang, Z. A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load. Mech. Syst. Signal Process. 2018, 100, 439–453. [Google Scholar] [CrossRef]
- Sainath, T.N.; Kingsbury, B.; Saon, G.; Soltau, H.; Mohamed, A.R.; Dahl, G.; Ramabhadran, B. Deep Convolutional Neural Networks for Large-scale Speech Tasks. Neural Netw. 2015, 64, 39–48. [Google Scholar] [CrossRef] [PubMed]
- Lai, P.R.; Wang, J.S. Multi-stage Attention Convolutional Neural Networks for HEVC In-Loop Filtering. In Proceedings of the 2020 2nd IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), Genova, Italy, 31 August–2 September 2020. [Google Scholar]
- Chen, W.; Shi, K. Multi-scale Attention Convolutional Neural Network for time series classification. Neural Netw. 2021, 136, 126–140. [Google Scholar] [CrossRef] [PubMed]
- Jia, F.; Lei, Y.; Lu, N.; Xing, S. Deep normalized convolutional neural network for imbalanced fault classification of machinery and its understanding via visualization. Mech. Syst. Signal Process. 2018, 110, 349–367. [Google Scholar] [CrossRef]
- Yang, B.; Lei, Y.; Jia, F.; Xing, S. An intelligent fault diagnosis approach based on transfer learning from laboratory bearings to locomotive bearings. Mech. Syst. Signal Process. 2019, 122, 692–706. [Google Scholar] [CrossRef]
- Sun, R.Y. Optimization for Deep Learning: An Overview. J. Oper. Res. Soc. China 2020, 3, 1–46. [Google Scholar] [CrossRef]
- Laurens, V.D.M.; Hinton, G. Visualizing Data using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579–2605. [Google Scholar]
- Shao, H.; Jiang, H.; Zhang, H.; Duan, W.; Liang, T.; Wu, S. Rolling bearing fault feature learning using improved convolutional deep belief network with compressed sensing. Mech. Syst. Signal Process. 2018, 100, 743–765. [Google Scholar] [CrossRef]
- Wang, X.; Wang, J.; Privault, M. Artificial intelligent fault diagnosis system of complex electronic equipment. J. Intell. Fuzzy Syst. 2018, 35, 4141–4151. [Google Scholar] [CrossRef]
- Zhu, X.; Luo, X.; Ye, X.; Han, Z.; Liu, Y. State Recognition Method of Turbine Rotor Based on Depth Feature Learning. Proc. CSEE 2021, 41, 432–442. [Google Scholar]
- Wen, L.; Li, X.; Gao, L.; Zhang, Y. A New Convolutional Neural Network-Based Data-Driven Fault Diagnosis Method. IEEE Trans. Ind. Electron. 2018, 65, 5990–5998. [Google Scholar] [CrossRef]
- Qu, J.; Yu, L.; Yuan, T.; Tian, Y. Adaptive fault diagnosis algorithm for rolling bearings based on one-dimensional convolutional neural network. Yi Qi Yi Biao Xue Bao/Chin. J. Sci. Instrum. 2018, 39, 134–143. [Google Scholar]
Sample Type | Sampling Time (s) | Load (NM) | Rotation Speed (rpm) | Sampling Frequency (Hz) |
---|---|---|---|---|
Normal | 60 | 50 | 1200 | 2 k/4 k/8 k/16 k/32 k |
1500 | 2 k/4 k/8 k/16 k/32 k | |||
1800 | 2 k/4 k/8 k/16 k/32 k | |||
2100 | 2 k/4 k/8 k/16 k/32 k |
Sample Type | Sample Length | Sample Size | |
---|---|---|---|
Train | Test | ||
Normal | 10,240 | 300 | 100 |
Broken gear tooth | 10,240 | 300 | 100 |
Gear crack | 10,240 | 300 | 100 |
Gear wear and tear | 10,240 | 300 | 100 |
Gear tooth deficiency | 10,240 | 300 | 100 |
Layer | Kernels/Filter (Height × Width/Stride) | Kernel Channel Size | Feature Maps |
---|---|---|---|
Input | 1 × 10,240 | ||
Convolution1 | 32 | 1 × 10,240 × 32 | |
Convolution2 | 32 | 1 × 10,240 × 32 | |
Max pooling3 | 1 × 2/2 | 32 | 1 × 5120 × 32 |
Convolution4 | 64 | 1 × 5120 × 64 | |
Convolution5 | 64 | 1 × 5120 × 64 | |
Max pooling6 | 1 × 2/2 | 64 | 1 × 2560 × 64 |
Fully connected | 64 | 1 | 64 × 1 |
Output | 5 | 1 | 5 |
C4, C5 | 1 × 1 | 1 × 2 | 1 × 4 | 1 × 6 | 1 × 8 | |
---|---|---|---|---|---|---|
C1, C2 | ||||||
1 × 1 | 57.6% | 63.4% | 52.3% | 57.8% | 55% | |
1 × 2 | 85.3% | 85.5% | 80.1% | 75.8% | 77.2% | |
1 × 4 | 80% | 77.2% | 85.6% | 85.5% | 83.1% | |
1 × 6 | 85.2% | 88.3% | 90.9% | 82.2% | 82.5% | |
1 × 8 | 82.9% | 80% | 88.5% | 96.3% | 82.6% |
Layer | Kernels/Filter (Height × Width/Stride) | Kernel Channel Size | Feature Maps |
---|---|---|---|
Input | 1 × 10,240 | ||
Convolution1 | 1 × 8/1 | 32 | 1 × 10,240 × 32 |
Convolution2 | 1 × 8/1 | 32 | 1 × 10,240 × 32 |
Max pooling3 | 1 × 2/2 | 32 | 1 × 5120 × 32 |
Convolution4 | 1 × 6/1 | 64 | 1 × 5120 × 64 |
Convolution5 | 1 × 6/1 | 64 | 1 × 5120 × 64 |
Max pooling6 | 1 × 2/2 | 64 | 1 × 2560 × 64 |
Fully connected | 64 | 1 | 64 × 1 |
Output | 5 | 1 | 5 |
Rotation Speed-Sampling Frequency | Recognition Rate | Std. |
---|---|---|
1200 rpm–2000 Hz | 96.3% | 0.034 |
1200 rpm–4000 Hz | 91% | 0.128 |
1200 rpm–8000 Hz | 83.2% | 0.113 |
1200 rpm–16,000 Hz | 76.8% | 0.265 |
1200 rpm–32,000 Hz | 71% | 0.227 |
Rotation Speed-Sampling Frequency | Recognition Rate | Std. |
---|---|---|
1200 rpm–2000 Hz | 96.3% | 0.034 |
1500 rpm–2000 Hz | 91.7% | 0.082 |
1800 rpm–2000 Hz | 88.1% | 0.065 |
2100 rpm–2000 Hz | 86.5% | 0.144 |
Rotation Speed-Sampling Frequency | Recognition Rate | Std. |
---|---|---|
1200 rpm–2000 Hz | 96.3% | 0.034 |
1200 rpm–4000 Hz → 2000 Hz | 93.5% | 0.058 |
1200 rpm–8000 Hz → 2000 Hz | 93.2% | 0.022 |
1200 rpm–16,000 Hz → 2000 Hz | 91.8% | 0.036 |
1200 rpm–32,000 Hz → 2000 Hz | 91.3% | 0.089 |
Rotation Speed-Sampling Frequency | Recognition Rate | Std. |
---|---|---|
1200 rpm–2000 Hz | 96.3% | 0.034 |
1500 rpm–2500 Hz | 96.7% | 0.027 |
1800 rpm–3000 Hz | 97.4% | 0.046 |
2100 rpm–3500 Hz | 98% | 0.013 |
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
Zhu, X.; Liu, B.; Li, Z.; Lin, J.; Gao, X. Research on Deep Learning Method and Optimization of Vibration Characteristics of Rotating Equipment. Sensors 2022, 22, 3693. https://doi.org/10.3390/s22103693
Zhu X, Liu B, Li Z, Lin J, Gao X. Research on Deep Learning Method and Optimization of Vibration Characteristics of Rotating Equipment. Sensors. 2022; 22(10):3693. https://doi.org/10.3390/s22103693
Chicago/Turabian StyleZhu, Xiaoxun, Baoping Liu, Zhentao Li, Jiawei Lin, and Xiaoxia Gao. 2022. "Research on Deep Learning Method and Optimization of Vibration Characteristics of Rotating Equipment" Sensors 22, no. 10: 3693. https://doi.org/10.3390/s22103693