A Novel Method for Identifying Crack and Shaft Misalignment Faults in Rotor Systems under Noisy Environments Based on CNN
Abstract
:1. Introduction
2. Introduction of Convolutional Neural Network
2.1. Convolutional Layer
2.2. Batch Normalization Layer
2.3. Pooling Layer
2.4. Fully Connected Layer
3. The Proposed Intelligent Diagnosis Method
- (1)
- When the training samples are original measured signals, the GA is used to optimize the main structural parameters of the CNN, and the batch size and learning rate of which are adjusted to adapt to the dataset. The raw time-domain signals are directly used as the input of the optimized CNN, then the noisy signals with different SNR values are used to verify the performance of the CNN model.
- (2)
- In the actual industrial environment, the complicated structure of large-scale mechanical equipment and the interaction between different components make it difficult to obtain the original vibration signals, and in the absence of a sample database, it is necessary to use the noisy signals in network training; on the other hand, the vibration signals measured on a test rig are invulnerable to noise, while signals in engineering have relatively low SNR values. In this paper, vibration signals in an actual noisy environment are simulated by adding noise to the vibration signals measured on a test rig, and a fault identification method combining VMD, PPCA, and CNN is proposed to achieve high-precision identification of different types of faults for a rotor system in a noisy environment.
3.1. Parameter Optimization of CNN Based on GA
3.2. VMD
3.3. PPCA
3.4. General Procedure of the Proposed Method
- Step 1: Different operating conditions of the rotor system are implemented in the test rig, and the vibration signals of the rotor system are collected by eddy current displacement sensors and a data acquisition system.
- Step 2: The collected raw time-domain signals are randomly divided into a training set and a testing set. The training set is directly input to a CNN for training, and the testing set is contaminated with Gaussian white noise to verify the performance of the CNN model. At the same time, the GA is used to optimize the numbers of convolutional layers, convolutional kernels, fully connected layers, and nodes in fully connected layers to obtain the best network structure parameters.
- Step 3: The batch size and learning rate are further determined to obtain the optimal network model.
- Step 4: Noises with different energy levels are added to the collected vibration signals, and the noisy signals are decomposed into k submodal functions with different frequency bands by VMD algorithm.
- Step 5: The decomposed k-dimensional signals are reduced to two-dimensional ones by PPCA to realize the separation of the denoised useful signals and the noise.
- Step 6: The denoised useful signals are taken as the input of the optimized CNN model, and the fault identification of crack and shaft misalignment for the rotor system in different noisy environments are realized through network learning.
4. Experimental Verification
4.1. Experimental Setup
4.2. Data Description
- (1)
- Adjust the motor control system to a specific rotating speed and keep the speed for at least 60 s. Then, collect the vibration signals for 20 s by the data acquisition system with a sampling frequency of 5 kHz;
- (2)
- Change the motor speed and repeat the vibration signal acquisition as mentioned above;
- (3)
- Change the fault types of the rotor system and repeat the above experimental operations.
4.3. Network Construction of CNN
4.3.1. Parameter Optimization Based on GA
4.3.2. Small Batch Size Training and Learning Rate Decay
4.4. Case 1: Fault Diagnosis for Noise-Free Training Samples
4.5. Case 2: Fault Diagnosis for Noise-Added Training Samples
4.5.1. Results and Discussions
4.5.2. Comparison of the Proposed Method and Other Methods
4.5.3. Verification Results at Different Rotating Speeds
4.5.4. Visualization of Network Learning
5. Conclusions
- (1)
- The numbers of convolutional layers, kernels, fully connected layers, and nodes in the fully connected layers can be determined automatically by GA, reducing the manual adjustment process which often depends on expertise and is time-consuming. The optimized CNN has excellent domain adaptability and can effectively identify multitype faults in rotor systems.
- (2)
- VMD and PPCA can not only remove the noise from vibration signals, but also preserve and amplify feature information in signals; combined with the optimized CNN, fault identification of crack and shaft misalignment for rotor systems under noisy environments can be effectively achieved.
- (3)
- The recognition rates for the faults of crack and shaft misalignment in the tested rotor system with different SNR values and rotating speeds all reached more than 99%.
Author Contributions
Funding
Conflicts of Interest
References
- Zhao, J.; DeSmidt, H.; Peng, M. Harmonic transfer function based damage identification of breathing cracked Jeffcott rotor. Shock Vib. 2016. [Google Scholar] [CrossRef]
- Bieryla, D.J.; Trethewey, M.W.; Lissenden, C.J.; Lebold, M.S.; Mitchell, S.; Maynard, K.P. Shaft crack monitoring via torsional vibration analysis. Part 1–Laboratory tests. In Proceedings of the 23nd International Modal Analysis Conference, Orlando, FL, USA, 31 January–3 February 2005. [Google Scholar]
- Rahman, A.G.A.; Ismail, Z.; Noroozi, S.; Chao, O.Z. Study of open crack in rotor shaft using changes in frequency response function phase. Int. J. Damage Mech. 2013, 22, 791–807. [Google Scholar] [CrossRef]
- Tlaisi, A.; Akinturk, A.; Swamidas, A.S.J.; Haddara, M.R. Crack detection in shaft using lateral and torsional vibration measurements and analyses. Turbomachinery 2012, 7, 693. [Google Scholar]
- Ishida, Y.; Inoue, T. Detection of a rotor crack using a harmonic excitation and nonlinear vibration analysis. J. Vib. Acoust. 2006, 12, 741. [Google Scholar] [CrossRef]
- Mani, G.; Quinn, D.D.; Kasarda, M.E.F. Active health monitoring in a rotating cracked shaft using active magnetic bearings as force actuators. J. Sound Vib. 2006, 294, 454–465. [Google Scholar] [CrossRef]
- Mani, G.; Quinn, D.D.; Kasarda, M.E.F. Structural health monitoring of rotor dynamic systems by wavelet analysis. In Proceedings of the ASME International Mechanical Engineering Congress and Exposition, American Society of Mechanical Engineers, Chicago, IL, USA, 5–10 November 2006. [Google Scholar]
- Chasalevris, A.C.; Papadopoulos, C.A. Early detection of rotor cracks by measuring the coupled response under external excitation. In Proceedings of the 8th IFToMM International Conference on Rotor Dynamics, Seoul, Korea, 15–20 September 2010; pp. 777–784. [Google Scholar]
- Sinha, J.K. Higher order spectra for crack and misalignment identification in the shaft of a rotating machine. Struct. Health Monit. 2007, 6, 325–334. [Google Scholar] [CrossRef]
- Sinha, J.K. Higher Order Coherences for fatigue crack detection. Eng. Struct. 2009, 31, 534–538. [Google Scholar] [CrossRef]
- Guo, D.; Peng, Z.K. Vibration analysis of a cracked rotor using Hilbert–Huang transform. Mech. Syst. Signal Process. 2007, 21, 3030–3041. [Google Scholar] [CrossRef]
- Babu, T.R.; Srikanth, S.; Sekhar, A.S. Hilbert–Huang transform for detection and monitoring of crack in a transient rotor. Mech. Syst. Signal Process. 2008, 22, 905–914. [Google Scholar] [CrossRef]
- Nagaraju, C.; Rao, K.N.; Rao, K.M. Application of 3D wavelet transforms for crack detection in rotor systems. Sadhana 2009, 34, 407–419. [Google Scholar] [CrossRef]
- Jesus, G.M.; Corral, E.; Castejon, C.; Garcia-Prada, J.C. Effective crack detection in railway axles using vibration signals and WPT energy. Sensors 2018, 18, 1603. [Google Scholar]
- Saridakis, K.M.; Chasalevris, A.C.; Papadopoulos, C.A.; Dentsoras, A.J. Applying neural networks, genetic algorithms and fuzzy logic for the identification of cracks in shafts by using coupled response measurements. Comput. Struct. 2008, 86, 1318–1338. [Google Scholar] [CrossRef]
- Sinou, J.J. Detection of cracks in rotor based on the 2× and 3× super-harmonic frequency components and the crack–unbalance interactions. Commun. Nonlinear Sci. Numer. Simul. 2008, 13, 2024–2040. [Google Scholar] [CrossRef]
- Guo, C.Z.; Yan, J.H.; Yang, W.C. Crack detection for a Jeffcott rotor with a transverse crack: An experimental investigation. Mech. Syst. Signal Process. 2017, 83, 260–271. [Google Scholar] [CrossRef]
- Jun, O.; Eun, H.; Earmme, Y.; Lee, C.-W. Modeling and vibration analysis of a simple rotor with a breathing crack. J. Sound Vib. 1992, 155, 273–290. [Google Scholar] [CrossRef]
- Darpe, A.; Gupta, K.; Chawla, A. Coupled bending, longitudinal and torsional vibrations of a cracked rotor. J. Sound Vib. 2004, 269, 33–60. [Google Scholar] [CrossRef]
- Darpe, A.K. Coupled vibrations of a rotor with slant crack. J. Sound Vib. 2007, 305, 172–193. [Google Scholar] [CrossRef]
- Saavedra, P.N.; Cuitiño, L.A. Vibration analysis of rotor for crack identification. J. Vib. Control 2002, 8, 51–67. [Google Scholar] [CrossRef]
- Prabhakar, S.; Sekhar, A.S.; Mohanty, A.R. Crack versus coupling misalignment in a transient rotor system. J. Sound Vib. 2002, 256, 773–786. [Google Scholar] [CrossRef]
- Patel, T.H.; Darpe, A.K. Vibration response of misaligned rotors. J. Sound Vib. 2009, 325, 609–628. [Google Scholar] [CrossRef]
- Sekhar, A.S.; Prabhu, B.S. Effects of coupling misalignment on vibrations of rotating machinery. J. Sound Vib. 1995, 185, 655–671. [Google Scholar] [CrossRef]
- Ioffe, S.; Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In Proceedings of the International Conference on Machine Learning, Lille, France, 6–11 July 2015. [Google Scholar]
- Dragomiretskiy, K.; Zosso, D. Variational mode decomposition. IEEE Trans. Signal Process. 2014, 62, 531–544. [Google Scholar] [CrossRef]
- Rockafellar, R.T. A dual approach to solving nonlinear programming problems by unconstrained optimization. Math. Program 1973, 5, 354–373. [Google Scholar] [CrossRef]
- Tipping, M.E.; Bishop, C.M. Probabilistic principal component analysis. J. R. Stat. Soc. Ser. B Stat. Methodol. 1999, 61, 611–622. [Google Scholar] [CrossRef]
- Utsugi, A.; Kumagai, T. Bayesian analysis of mixtures of factor analyzers. Neural Comput. 2001, 13, 993–1002. [Google Scholar] [CrossRef]
- Cardoso, F.R.; Achcar, J.A.; Piratelli, C.L.; Hermosilla, J.L.G.; Barbosa, J.C. Bayesian analysis of employee suggestions in a food company. Int. J. Adv. Manuf. Technol. 2014, 70, 2059–2070. [Google Scholar] [CrossRef]
- Dempster, A.P.; Laird, N.M.; Rubin, D.B. Maximum likelihood from incomplete data via the EM algorithm. J. R. Stat. Soc. Ser. B Stat. Methodol. 1977, 39, 1–38. [Google Scholar]
- Jordan, M.I.; Jacobs, R.A. Hierarchical mixtures of experts and the EM algorithm. Neural Comput. 1994, 6, 181–214. [Google Scholar] [CrossRef]
- McLachlan, G.; Krishnan, T. The EM Algorithm and Extensions, 2nd ed.; Wiley: New York, NY, USA, 2008. [Google Scholar]
- Zhang, W.; Peng, G.L.; Li, C.H.; Chen, Y.H.; Zhang, Z.J. A new deep learning model for fault diagnosis with good anti-noise and domain adaptation ability on raw vibration signals. Sensors 2017, 17, 452. [Google Scholar] [CrossRef]
- Goyal, P.; Dollar, P.; Girshick, R.B.; Noordhuis, P.; Wesolowski, L.; Kyrola, A. Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour. arXiv 2017, arXiv:1706.02677. [Google Scholar]
- Keskar, N.S.; Mudigere, D.; Nocedal, J.; Smelyanskiy, M.; Tang, P.T.P. On large-batch training for deep learning: Generalization gap and sharp minima. In Proceedings of the 5th International Conference on Learning Representations, Toulon, France, 24–26 April 2016. [Google Scholar]
- Wang, Y.X.; Markert, R. Filter bank property of variational mode decomposition and its applications. IEEE Trans. Signal Process. 2016, 120, 509–521. [Google Scholar] [CrossRef]
- Li, J.; Yao, X.; Wang, H.; Zhang, J. Periodic impulses extraction based on improved adaptive VMD and sparse code shrinkage denoising and its application in rotating machinery fault diagnosis. Mech. Syst. Signal Process. 2019, 126, 568–589. [Google Scholar] [CrossRef]
- Zhou, C.J.; Ma, J.; Wu, J.D.; Yuan, X.Y. An adaptive VMD method based on improved GOA to extract early fault feature of rolling bearings. Int. J. Innov. Comput. Inf. Control 2019, 15, 1485–1505. [Google Scholar]
- Yi, C.C.; Lv, Y.; Dang, Z. A fault diagnosis scheme for rolling bearing based on particle swarm optimization in variational mode decomposition. Shock Vib. 2016. [Google Scholar] [CrossRef]
- Wang, X.B.; Yang, Z.X.; Yan, X.A. Novel particle swarm optimization-based variational mode decomposition method for the fault diagnosis of complex rotating machinery. IEEE-ASME Trans. Mech. 2018, 23, 68–79. [Google Scholar] [CrossRef]
- Lian, J.J.; Liu, Z.; Wang, H.J.; Dong, X.F. Adaptive variational mode decomposition method for signal processing based on mode characteristic. Mech. Syst. Signal Process. 2018, 107, 53–77. [Google Scholar] [CrossRef]
- Li, Z.P.; Chen, J.L.; Zi, Y.Y.; Pan, J. Independence-oriented VMD to identify fault feature for wheel set bearing fault diagnosis of high speed locomotive. Mech. Syst. Signal Process. 2017, 85, 512–529. [Google Scholar] [CrossRef]
- Maaten, L.V.D.; Hinton, G. Visualizing data using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579–2605. [Google Scholar]
Components | Type | Components | Type |
---|---|---|---|
Driving motor | SIEMENSE 1.5 kW | Coupling | Flexible coupling |
Motor control system | VFD-M 1.5 kW | Bearing | SKF 6300 |
Eddy current displacement sensor | ZA21-0803 | The material of rotary shaft | 40Cr |
Signal acquisition system | DHDAS5922N | The material of rotary disc | 45# steel |
Description | Symbol | Value |
---|---|---|
Population size | N | 50 |
Maximum generation number | n | 40 |
The probability of crossover | 0.95 | |
The probability of mutation | 0.2 | |
Maximum number of convolutional layers | 5 | |
Maximum number of fully connected layers | 3 | |
Maximum number of kernels in convolutional layers | 64 | |
Maximum number of nodes in fully connected layers | 512 |
Parameter | Value |
---|---|
The number of convolutional layers | 4 |
The number of kernels in conv1 | 8 |
The number of kernels in conv2 | 16 |
The number of kernels in conv3 | 32 |
The number of kernels in conv4 | 32 |
The number of fully connected layers | 1 |
The number of nodes in fully connected layer | 128 |
SNR (dB) | −10 | −6 | −2 | 0 | 2 | 6 | 10 |
---|---|---|---|---|---|---|---|
Accuracy (%) | 87.67 | 99.83 | 100 | 100 | 100 | 100 | 100 |
SNR (dB) | −10 | −6 | −2 | 0 | 2 | 6 | 10 |
---|---|---|---|---|---|---|---|
Accuracy (%) | 99.83 | 99.92 | 99.25 | 100 | 100 | 99.92 | 100 |
© 2019 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 (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Zhao, W.; Hua, C.; Dong, D.; Ouyang, H. A Novel Method for Identifying Crack and Shaft Misalignment Faults in Rotor Systems under Noisy Environments Based on CNN. Sensors 2019, 19, 5158. https://doi.org/10.3390/s19235158
Zhao W, Hua C, Dong D, Ouyang H. A Novel Method for Identifying Crack and Shaft Misalignment Faults in Rotor Systems under Noisy Environments Based on CNN. Sensors. 2019; 19(23):5158. https://doi.org/10.3390/s19235158
Chicago/Turabian StyleZhao, Wang, Chunrong Hua, Dawei Dong, and Huajiang Ouyang. 2019. "A Novel Method for Identifying Crack and Shaft Misalignment Faults in Rotor Systems under Noisy Environments Based on CNN" Sensors 19, no. 23: 5158. https://doi.org/10.3390/s19235158
APA StyleZhao, W., Hua, C., Dong, D., & Ouyang, H. (2019). A Novel Method for Identifying Crack and Shaft Misalignment Faults in Rotor Systems under Noisy Environments Based on CNN. Sensors, 19(23), 5158. https://doi.org/10.3390/s19235158