Multi-Stream Convolutional Neural Networks for Rotating Machinery Fault Diagnosis under Noise and Trend Items
<p>Image of time-domain input (TD).</p> "> Figure 2
<p>Image of FFT-based frequency-domain input (FD).</p> "> Figure 3
<p>Image of STFT-based time–frequency domain input (STFT-TFD).</p> "> Figure 4
<p>Image of WT-based time–frequency domain input (WT-TFD).</p> "> Figure 5
<p>Architecture of the residual block.</p> "> Figure 6
<p>Two examples of multi-stream networks: (<b>a</b>) the structure fusing information after Conv and (<b>b</b>) the structure fusing information after Fc.</p> "> Figure 7
<p>Structure of the final proposed network.</p> "> Figure 8
<p>Performance comparisons: (<b>a</b>) tests on original datasets, (<b>b</b>) tests on datasets with noise, and (<b>c</b>) tests on datasets with trend items.</p> "> Figure 9
<p>FLOPs comparisons.</p> "> Figure 10
<p>Feature maps visualization for CWRU with noise and trend items.</p> "> Figure 11
<p>Feature maps visualization for UoC with noise and trend items: (<b>a</b>) visualization of FD-stream before the first fusion, (<b>b</b>) visualization of TFD-stream before the first fusion, (<b>c</b>) visualization of TD-stream before the first fusion, and (<b>d</b>) visualization after the second fusion.</p> "> Figure 12
<p>Feature maps visualization for SEU with noise and trend items: (<b>a</b>) visualization of FD-stream before the first fusion, (<b>b</b>) visualization of TFD-stream before the first fusion, (<b>c</b>) visualization of TD-stream before the first fusion, and (<b>d</b>) visualization after the second fusion.</p> ">
Abstract
:1. Introduction
- (1)
- Based on three public datasets, we conduct comprehensive experiments on four input types: time domain input, frequency domain input, STFT-based time–frequency input, and WT-based time–frequency input, with networks of three depths, two input sizes, and three types of interfering signals. Through the experiments, the suitable neural network depths and image sizes for these four input types are further obtained.
- (2)
- Through theoretical analysis as well as analysis of experimental results, we study the difference in characteristics between the four input types, including the carried information, robustness to noise and trend items, learning difficulty for CNN models, etc. It preliminarily demonstrates the complementarity of information between different input types.
- (3)
- We design a series of fusion models and conduct experiments to investigate where and how to fuse the three networks. Based on this, we proposed the final multi-stream convolutional neural network, which performs well under different environments, without any data pre-cleaning.
- (4)
- We try to explore the inner mechanism of the proposed model by visualizing the learned feature maps. The feature distributions learned by different streams are different, which further demonstrates the complementarity of information. The fusion layers can fuse the input features well and help improve the classification ability of the network.
2. Input Types Definition and Discussion
2.1. Input Types Definition
2.1.1. Time-Domain Input
2.1.2. Frequency-Domain Input
2.1.3. STFT Based Time–Frequency Domain Input
2.1.4. WT-Based Time–Frequency Domain Input
2.2. Input Types Discussion
3. CNN-Based Fault Diagnosis Evaluations
3.1. Datasets
3.1.1. CWRU Bearing Datasets
3.1.2. UoC Gear Fault Datasets
3.1.3. SEU Gearbox Datasets
3.2. CNN Models
3.3. Evaluations without Interfering Signals
3.4. Evaluations with Interfering Signals
- (1)
- TD is significantly affected by input size and data size, and it always performs better at higher resolutions and larger data sizes. It can be shown from Table 10 that TD performs the worst with noise compared to the other three input types containing frequency domain information, which means its poor robustness to noise. By comparing the performance of TD in different situations, it can be seen that TD is almost unaffected by the trend items. Surprisingly, in Table 12, the performance of TD on 50% SEU and 100% SEU is the best among the four input types, indicating that TD contains rich health information but is difficult to train and hard to fit.
- (2)
- As can be seen from Table 11 and Table 12, FD requires higher resolution to achieve better prediction accuracy when the interfering signal contains trend items, but it is less affected by image resolution than TD. In most cases, FD performs best with interference containing only noise, indicating its great robustness to noise. However, FD is less robust to trend items relative to TD, which should be due to the fact that the frequencies of trend items may mask the main frequencies of high frequencies.
- (3)
- STFT-TFD is insensitive to resolution and often performs better at lower resolution. STFT-TFD has excellent robustness to trend items, almost unaffected, and its robustness to noise is also good.
- (4)
- WT-TFD is insensitive to resolution, and its performance is greatly affected by data scale. WT-TFD is more robust to noise than TD, but it performs poorly under trend items.
4. Proposed Method
4.1. Where and How to Fuse the Streams
4.2. Multi-Stream Convolutional Neural Network
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Lee, J. Industrial Big Data; China Machine Press: Beijing, China, 2015; Chapter 2. [Google Scholar]
- Lei, Y.; Yang, B.; Jiang, X.; Jia, F.; Li, N.; Nandi, A.K. Applications of machine learning to machine fault diagnosis: A review and roadmap. Mech. Syst. Signal Process. 2020, 138, 106587. [Google Scholar] [CrossRef]
- Shao, S.; McAleer, S.; Yan, R.; Baldi, P. Highly-accurate machine fault diagnosis using deep transfer learning. IEEE Trans. Ind. Inf. 2019, 15, 2446–2455. [Google Scholar] [CrossRef]
- Verstraete, D.; Ferrada, A.; Droguett, E.L.; Meruane, V.; Modarres, M. Deep learning enabled fault diagnosis using time-frequency image analysis of rolling element bearings. Shock Vib. 2017, 2017, 5067651. [Google Scholar] [CrossRef]
- LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444. [Google Scholar] [CrossRef]
- Jiao, J.; Zhao, M.; Lin, J.; Liang, K. A comprehensive review on convolutional neural network in machine fault diagnosis. Neurocomputing 2020, 417, 36–63. [Google Scholar] [CrossRef]
- Han, Y.; Tang, B.; Deng, L. Multi-level wavelet packet fusion in dynamic ensemble convolutional neural network for fault diagnosis. Measurement 2018, 417, 246–255. [Google Scholar] [CrossRef]
- Janssens, O.; Slavkovikj, V.; Vervisch, B.; Stockman, K.; Loccufier, M.; Verstockt, S.; van de Walle, R.; van Hoecke, S. Convolutional neural network based fault detection for rotating machinery. J. Sound Vib. 2016, 377, 331–345. [Google Scholar] [CrossRef]
- Cao, P.; Zhang, S.; Tang, J. Preprocessing-Free Gear Fault Diagnosis Using Small Datasets with Deep Convolutional Neural Network-Based Transfer Learning. IEEE Access 2018, 6, 26241–26253. [Google Scholar] [CrossRef]
- Xin, Y.; Li, S.; Cheng, C.; Wang, J. An intelligent fault diagnosis method of rotating machinery based on deep neural networks and time-frequency analysis. J. Vibroeng. 2018, 20, 2321–2335. [Google Scholar] [CrossRef]
- Ma, S.; Chu, F.; Han, Q. Deep residual learning with demodulated time-frequency features for fault diagnosis of planetary gearbox under nonstationary running conditions. Mech. Syst. Signal Process. 2019, 127, 190–201. [Google Scholar] [CrossRef]
- Jin, Y.; Qin, C.; Huang, Y.; Liu, C. Actual bearing compound fault diagnosis based on active learning and decoupling attentional residual network. Measurement 2021, 173, 108500. [Google Scholar] [CrossRef]
- Guo, S.; Yang, T.; Gao, W.; Zhang, C.; Zhang, Y. An intelligent fault diagnosis method for bearings with variable rotating speed based on Pythagorean spatial pyramid pooling CNN. Sensors 2018, 18, 3857. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Zhao, M.; Kang, M.; Tang, B.; Pecht, M. Deep residual networks with dynamically weighted wavelet coefficients for fault diagnosis of planetary gearboxes. IEEE Trans. Ind. Electron. 2018, 65, 4290–4300. [Google Scholar] [CrossRef]
- Zhao, M.; Kang, M.; Tang, B.; Pecht, M. Multiple wavelet coefficients fusion in deep residual networks for fault diagnosis. IEEE Trans. Ind. Electron. 2019, 66, 4696–4706. [Google Scholar] [CrossRef]
- Cao, X.; Chena, B.; Yaoa, B.; He, W. Combining translation-invariant wavelet frames and convolutional neural network for intelligent tool wear state identification. Comput. Ind. 2019, 106, 71–84. [Google Scholar] [CrossRef]
- Theodoropoulos, P.; Spandonidis, C.C.; Fassois, S. Use of Convolutional Neural Networks for vessel performance optimization and safety enhancement. Ocean Eng. 2022, 248, 110771. [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]
- Dybała, J.; Zimroz, R. Application of empirical mode decomposition for impulsive signal extraction to detect bearing damage-industrial case study. In Proceedings of the Condition Monitoring of Machinery in Non-Stationnary Operations, Hammamet, Tunisia, 26–28 March 2012; pp. 257–266. [Google Scholar]
- Dybała, J.; Zimroz, R. Rolling bearing diagnosing method based on Empirical Mode Decomposition of machine vibration signal. Appl. Acoust. 2014, 77, 195–203. [Google Scholar] [CrossRef]
- Bin, G.F.; Liao, C.J.; Li, X.J. The method of fault feature extraction from acoustic emission signals using Wigner-Ville distribution. In Proceedings of the International Conference on Optical, Electronic Materials and Applications, Chongqing, China, 4–6 March 2011; pp. 732–737. [Google Scholar]
- Han, H.; Hao, L. Fault diagnosis method of rolling bearings Based on SPA-FE-IFSVM. Adv. Mech. Eng. 2020, 12, 1687814020969470. [Google Scholar] [CrossRef]
- Bagherzadeh, S.A.; Salehi, M. Analysis of in-flight cabin vibration of a turboprop airplane by proposing a novel noise-tolerant signal decomposition method. J. Vib. Control 2021, 1–14. [Google Scholar] [CrossRef]
- Wang, T.; Liang, M.; Li, J.; Cheng, W.; Li, C. Bearing fault diagnosis under unknown variable speed via gear noise cancellation and rotational order sideband identification. Mech. Syst. Signal Process. 2015, 62–63, 30–53. [Google Scholar] [CrossRef]
- Zhao, Z.; Li, T.; Wu, J.; Sun, C.; Wang, S.; Yan, R.; Chen, X. Deep learning algorithms for rotating machinery intelligent diagnosis: An open source benchmark study. ISA Trans. 2020, 107, 224–255. [Google Scholar] [CrossRef] [PubMed]
- Yan, X.; Jia, M. A novel optimized SVM classification algorithm with multi-domain feature and its application to fault diagnosis of rolling bearing. Neurocomputing 2018, 313, 47–64. [Google Scholar] [CrossRef]
- Li, C.; Sanchez, R.V.; Zurita, G.; Cerrada, M.; Cabrera, D.; Vásquez, R.E. Multimodal deep support vector classification with homologous features and its application to gearbox fault diagnosis. Neurocomputing 2015, 168, 119–127. [Google Scholar] [CrossRef]
- Lin, J.; Qu, L. Feature extraction based on morlet wavelet and its application for mechanical fault diagnosis. J. Sound Vib. 2000, 234, 135–148. [Google Scholar] [CrossRef]
- Lei, Y.; Zuo, M.J.; He, Z.; Zi, Y. A multidimensional hybrid intelligent method for gear fault diagnosis. Expert Syst. Appl. 2010, 37, 1419–1430. [Google Scholar] [CrossRef]
- Lei, Y.; He, Z.; Zi, Y.; Hu, Q. Fault diagnosis of rotating machinery based on multiple ANFIS combination with Gas. Mech. Syst. Signal Process. 2007, 21, 2280–2294. [Google Scholar] [CrossRef]
- Zhao, C.; Feng, Z. Application of multi-domain sparse features for fault identification of planetary gearbox. Measurement 2017, 104, 169–179. [Google Scholar] [CrossRef]
- Zhao, R.; Yan, R.; Chen, Z.; Mao, K.; Wang, P.; Gao, R.X. Deep learning and its applications to machine health monitoring. Mech. Syst. Signal Process. 2019, 115, 213–237. [Google Scholar] [CrossRef]
- Yun, W.; Li, C. Eliminating tendency processing methods of vibration signal based on MATLAB. Auto Eng. 2011, 6, 27–29. [Google Scholar]
- Case Western Reserve University. Bearing Data Center Website. Available online: http://csegroups.case.edu/bearingdatacenter/pages/download-data-file (accessed on 15 February 2022).
- Cao, P.; Zhang, S.; Tang, J. Gear Fault Data. Available online: https://doi.org/10.6084/m9.figshare.6127874.v1 (accessed on 15 February 2022).
- Shao, S.; McAleer, S.; Yan, R.; Baldi, P. SEU Gearbox Datasets. Available online: https://github.com/cathysiyu/Mechanical-datasets (accessed on 15 February 2022).
- Simonyan, K.; Zisserman, A. Very deep convolutional networks for large-scale image recognition. In Proceedings of the 3rd International Conference on Learning Representations, San Diego, CA, USA, 7–9 May 2015; pp. 1021–1026. [Google Scholar]
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the 29th Conference on Computer Vision and Pattern Recognition (CVPR 2016), Seattle, WA, USA, 1–26 July 2016; pp. 770–778. [Google Scholar]
- Xiao, L.; Ren, J.; Yang, H. Study on vibration signal pre-processing method based on MATLAB. Comput. Simul. 2010, 8, 330–337. [Google Scholar]
- Lian, J. Optimization of digital integral method for vibration signal under non-stationary condition. Mech. Electr. Eng. Technol. 2021, 50, 190–193. [Google Scholar]
- Hasan, M.J.; Islam, M.M.M.; Kim, J. Bearing fault diagnosis using multidomain fusion-based vibration imaging and multitask learning. Sensors 2022, 2, 56. [Google Scholar] [CrossRef] [PubMed]
- Molchanov, P.; Tyree, S.; Karras, T.; Aila, T.; Kautz, J. Pruning convolutional neural networks for resource efficient inference. In Proceedings of the 5th International Conference on Learning Representations (ICLR 2017), Toulon, France, 24–26 April 2017; pp. 1–17. [Google Scholar] [CrossRef]
- van der Maaten, L.; Hinton, G. Visualizing data using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579–2625. [Google Scholar]
Health State | Fault Position | Fault Diameter | Working Load | Sample Size |
---|---|---|---|---|
Health | - | - | 0/1/2/3 | 110/110/110/110 |
0.007Inner | inner race | 0.007 | 0/1/2/3 | 110/110/110/110 |
0.014Inner | inner race | 0.014 | 0/1/2/3 | 110/110/110/110 |
0.021Inner | inner race | 0.021 | 0/1/2/3 | 110/110/110/110 |
0.028Inner | inner race | 0.028 | 0/1/2/3 | 110/110/110/110 |
0.007Ball | rolling element | 0.007 | 0/1/2/3 | 110/110/110/110 |
0.014Ball | rolling element | 0.014 | 0/1/2/3 | 110/110/110/110 |
0.021Ball | rolling element | 0.021 | 0/1/2/3 | 110/110/110/110 |
0.028Ball | rolling element | 0.028 | 0/1/2/3 | 110/110/110/110 |
0.007Outer6 | outer race @6:00 | 0.007 | 0/1/2/3 | 110/110/110/110 |
0.007Outer3 | outer race @3:00 | 0.007 | 0/1/2/3 | 110/110/110/110 |
0.007Outer12 | outer race @12:00 | 0.007 | 0/1/2/3 | 110/110/110/110 |
0.014Outer6 | outer race @6:00 | 0.014 | 0/1/2/3 | 110/110/110/110 |
0.021Outer6 | outer race @6:00 | 0.021 | 0/1/2/3 | 110/110/110/110 |
0.021Outer3 | outer race @3:00 | 0.021 | 0/1/2/3 | 110/110/110/110 |
0.021Outer12 | outer race @12:00 | 0.021 | 0/1/2/3 | 110/110/110/110 |
Health State | Working Load | Sample Size |
---|---|---|
Health | 0/2 | 1000/1000 |
Chipped | 0/2 | 1000/1000 |
Miss | 0/2 | 1000/1000 |
Root | 0/2 | 1000/1000 |
Surface | 0/2 | 1000/1000 |
Ball | 0/2 | 1000/1000 |
Inner | 0/2 | 1000/1000 |
Outer | 0/2 | 1000/1000 |
Combination | 0/2 | 1000/1000 |
Layer Name | Output Size | ResNet18’ | ResNet34’ |
---|---|---|---|
conv1 | |||
conv2_x | |||
conv3_x | |||
conv4_x | |||
conv5_x | |||
average pool, fc, softmax |
Networks | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
VGG16 | 99.70% | 99.95% | 99.92% | 99.75% | |
99.60% | 99.95% | 100.0% | 99.75% | ||
ResNet18’ | 99.80% | 100.0% | 100.0% | 99.90% | |
99.50% | 99.95% | 100.0% | 99.85% | ||
ResNet34’ | 99.85% | 99.92% | 100.0% | 99.95% | |
99.55% | 99.95% | 100.0% | 99.95% |
Networks | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
VGG16 | 100.0% | 100.0% | 99.88% | 100.0% | |
99.88% | 99.88% | 100.0% | 99.88% | ||
ResNet18’ | 99.88% | 100.0% | 100.0% | 100.0% | |
99.65% | 100.0% | 100.0% | 100.0% | ||
ResNet34’ | 100.0% | 100.0% | 100.0% | 100.0% | |
100.0% | 100.0% | 100.0% | 100.0% |
Networks | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
VGG16 | 96.47% | 100.0% | 99.37% | 97.48% | |
95.96% | 99.41% | 98.91% | 98.65% | ||
ResNet18’ | 97.31% | 99.66% | 99.83% | 99.16% | |
96.04% | 99.50% | 99.37% | 99.16% | ||
ResNet34’ | 98.82% | 99.83% | 99.66% | 99.16% | |
96.30% | 99.92% | 99.33% | 98.91% |
Networks | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
VGG16 | 99.52% | 99.82% | 99.63% | 99.59% | |
97.67% | 99.82% | 99.48% | 99.37% | ||
ResNet18’ | 99.41% | 99.82% | 99.96% | 99.67% | |
97.74% | 99.82% | 99.74% | 99.52% | ||
ResNet34’ | 99.78% | 99.96% | 99.93% | 99.63% | |
98.30% | 99.85% | 99.59% | 99.33% |
Networks | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
VGG16 | 99.91% | 99.98% | 99.89% | 99.78% | |
99.20% | 100.0% | 99.82% | 99.85% | ||
ResNet18’ | 99.82% | 99.91% | 99.94% | 99.93% | |
99.04% | 99.93% | 99.82% | 99.96% | ||
ResNet34’ | 100.0% | 99.96% | 99.93% | 99.96% | |
99.22% | 99.94% | 99.91% | 99.93% |
Parameters | CWRU | UoC | 22%SEU | 50%SEU | 100%SEU |
---|---|---|---|---|---|
(1, 7) | (0.3, 0.7) | (0.01, 0.03) | |||
(1, 7) | (0.3, 0.7) | (0.01, 0.03) | |||
(1, 7) | (0.3, 0.7) | (0.01, 0.03) | |||
{1} | {0.1} | {0.01} | |||
{1} | {0.1} | {0.01} | |||
() | () | () | |||
() | () | () |
Datasets | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
CWRU | 94.14% | 97.63% | 96.77% | 96.11% | |
92.88% | 97.17% | 97.07% | 96.77% | ||
UoC | 94.92% | 95.51% | 96.81% | 95.27% | |
84.75% | 93.97% | 97.52% | 96.81% | ||
22%SEU | 91.75% | 95.37% | 93.69% | 93.18% | |
82.66% | 95.79% | 94.11% | 92.42% | ||
50%SEU | 94.85% | 96.96% | 94.26% | 94.41% | |
91.04% | 96.48% | 94.44% | 92.93% | ||
100%SEU | 96.48% | 97.82% | 95.57% | 96.13% | |
93.35% | 97.50% | 96.17% | 94.37% |
Datasets | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
CWRU | 99.85% | 99.75% | 99.90% | 89.65% | |
99.44% | 99.39% | 100.0% | 88.64% | ||
UoC | 100.0% | 99.17% | 100.0% | 97.05% | |
99.88% | 99.05% | 100.0% | 96.69% | ||
22%SEU | 99.33% | 98.65% | 99.50% | 96.89% | |
96.59% | 98.32% | 99.66% | 96.72% | ||
50%SEU | 99.85% | 99.52% | 99.83% | 97.85% | |
97.93% | 99.26% | 99.85% | 97.63% | ||
100%SEU | 99.96% | 99.80% | 99.76% | 98.70% | |
99.24% | 99.65% | 99.87% | 98.19% |
Datasets | Input Sizes | TD | FD | STFT-TFD | WT-TFD |
---|---|---|---|---|---|
CWRU | 94.34% | 96.01% | 93.69% | 85.15% | |
93.49% | 95.46% | 95.86% | 86.16% | ||
UoC | 94.56% | 91.73% | 93.74% | 90.90% | |
82.15% | 90.07% | 95.27% | 89.36% | ||
22%SEU | 88.13% | 90.66% | 90.15% | 88.90% | |
81.57% | 89.73% | 91.67% | 86.70% | ||
50%SEU | 94.78% | 94.44% | 92.67% | 92.11% | |
89.56% | 93.52% | 93.41% | 89.67% | ||
100%SEU | 96.00% | 95.67% | 94.83% | 93.78% | |
92.74% | 94.82% | 94.98% | 92.50% |
Datasets | Fusion Ways | Conv | Ap | Fc | Softmax |
---|---|---|---|---|---|
CWRU | sum | 96.87% | 97.53% | 97.32% | 97.68% |
cat | 96.97% | 97.12% | - | - | |
conv | 97.12% | 97.67% | 97.48% | - | |
UoC | sum | 97.52% | 97.40% | 97.28% | 97.99% |
cat | 97.40% | 96.57% | - | - | |
conv | 96.22% | 96.45% | 96.34% | - | |
22%SEU | sum | 94.53% | 94.70% | 94.78% | 95.37% |
cat | 93.27% | 93.35% | - | - | |
conv | 93.77% | 94.44% | 93.94% | - |
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
Dong, H.; Lu, J.; Han, Y. Multi-Stream Convolutional Neural Networks for Rotating Machinery Fault Diagnosis under Noise and Trend Items. Sensors 2022, 22, 2720. https://doi.org/10.3390/s22072720
Dong H, Lu J, Han Y. Multi-Stream Convolutional Neural Networks for Rotating Machinery Fault Diagnosis under Noise and Trend Items. Sensors. 2022; 22(7):2720. https://doi.org/10.3390/s22072720
Chicago/Turabian StyleDong, Han, Jiping Lu, and Yafeng Han. 2022. "Multi-Stream Convolutional Neural Networks for Rotating Machinery Fault Diagnosis under Noise and Trend Items" Sensors 22, no. 7: 2720. https://doi.org/10.3390/s22072720
APA StyleDong, H., Lu, J., & Han, Y. (2022). Multi-Stream Convolutional Neural Networks for Rotating Machinery Fault Diagnosis under Noise and Trend Items. Sensors, 22(7), 2720. https://doi.org/10.3390/s22072720