1. Introduction
With the development of industry, energy development has become crucial to driving social development. Energy is being extracted in large quantities and used in a crude manner. This has led to a gradual depletion of non-renewable energy sources. At the same time, the excessive use of fossil energy is causing increasing environmental pollution. This poses a serious threat to the sustainable development of the human economy and society. Countries around the world are actively developing policies on new energy sources in order to change the energy landscape in their countries. Wind energy is an important clean technology. Among the renewable energy technologies, wind power generation is a relatively mature and commercially promising option. The development of wind power is of great importance in improving the energy structure, protecting the ecological environment, ensuring energy security, and achieving sustainable economic development. It has become a worldwide consensus to vigorously develop wind power generation [
1,
2].
The wind turbine works in an alternating load environment. It is prone to failure. Most of the units are located in remote suburban areas, so manual inspections are costly, and routine maintenance is a major challenge. Rolling bearings are an important component of wind turbines. The health of rolling bearings is the basis for the stable operation of wind turbines. It is important to carry out research on the fault diagnosis of wind turbine bearings [
2,
3,
4].
Among the bearing fault diagnosis methods, there are two main categories: data-driven methods, and signal analysis methods [
5,
6,
7]. In terms of signal analysis methods, some recent research advances have been made. Wang Xiaolong et al. presented a diagnosis method based on the improved empirical wavelet transform (IEWT) [
8]. The constructed wavelet is used to separate the sensitive modal components from the angular domain signal. The transient energy amplification feature of the sensitive modal component is calculated by means of a frequency-weighted energy operator. The engineering field signal proves the effectiveness of this method. Wang Huaqing et al. presented a method to obtain comprehensive fault parameters using the least squares mapping (LSM) technique [
9]. This method can improve the diagnostic sensitivity of symptom parameters. Examples of bearing diagnostics verify that the method is effective in extracting fault characteristics such as outer raceways, inner raceways, and roller elements of bearings. This method can be used for fault diagnosis under the condition of variable speed. An EWT-MDS method was presented by Tan Yuan et al., it combines empirical wavelet decomposition and multidimensional scale transformation [
10]. In this method, the bearing signal is decomposed by a self-adaptive empirical wavelet. The change characteristics of each mode are analyzed by information entropy. By combining multidimensional scale transformation algorithms, the synergistic variation pattern of each component in high-dimensional space is obtained, thereby enabling bearing fault diagnosis. These methods rely on the engineering experience of technicians. It is difficult to make the related technologies universal. Especially under variable working conditions, the vibration signal has the characteristics of modulation, pulse impact interval, unstable sampling phase, and low signal-to-noise ratio. Fault diagnosis methods based on signal analysis are limited. Data-driven methods can significantly remove such limitations. Data-driven methods are based on the use of features inherent in the data for fault diagnosis studies. These methods can effectively and quickly process signals. Accurate detection results are provided. The serendipity of manually extracted features is avoided. Therefore, data-driven methods have recently been widely studied in the field of mechanical fault diagnosis.
Most data-driven methods are based on machine learning. As an important branch of machine learning, deep learning (DL) has recently been widely used in fault diagnosis. DL is an effective technique in data-driven methods, and it is different from one-dimensional vibration signal information. Such methods use the intrinsic characteristics of sample data for fault detection studies. Convolutional neural networks (CNNs) in DL are particularly suitable for processing two-dimensional images, due to their tight connections between levels in the spatial structure. They adaptively extract rich correlation properties between pixel points from the image. The layout of the CNN model is closer to an actual biological neural network than to an artificial neural network. Weight sharing reduces the complexity of the network. In the field of image processing, images can be used directly as inputs to a convolutional neural network model. The pattern recognition process of feature classification is avoided. Convolutional neural networks are more widely used than traditional neural networks. They are also playing a huge role in the rise of deep learning.
In terms of DL signal analysis methods, some recent research advances have been made. Chen et al. used continuous wavelet transform (CWT) to process the bearing vibration data into a wavelet time-frequency graph [
11]. Then, a square pool structure CNN was constructed to extract advanced features. The extreme learning machine was used as a strong classifier. The validity of the method was verified using a dataset of motor bearings. Pham et al. transformed the raw data into spectrograms using STFT. The spectrograms were sent to the neural network for fault classification [
12]. A high level of accuracy was achieved. Wang Nini et al. presented a rolling bearing fault diagnosis model based on multiscale deep convolution network feature fusion. A multiscale feature fusion module was built into the network structure to extract features at different levels of the fault sample. Precise classification of different faults was achieved [
13]. Wang et al. presented an image coding method based on a Gramian angular field (GAF). The validity of this method was verified by a tiled CNN. The GAF can maintain the time dependence of the signal vibration sequence and alter the original data distribution to make it easier to distinguish from Gaussian noise [
14].
CNNs have been used in mechanical fault diagnosis by a large number of researchers. In order to obtain better diagnostic accuracy, the designed neural network is often too deep and too wide, which brings about complex calculations and high memory consumption. Thus, the economic cost of data storage is increased. Applications in embedded mobile devices are limited. Therefore, on the basis of ResNet, inverted residual structure, and depth separable convolution, this paper presents the design of a double-layer separated residual convolution module as the main structure [
15,
16,
17]. The multi-branch structure is used as the initial information-receiving domain of the sample. In order to enhance the nonlinear expression ability of the model, efficient channel attention (ECA) is introduced into the structure to capture the local features in information transmission [
18]. A double-layer separation residual CNN (DRCNN) model with small storage, low latency, and high accuracy is constructed. Firstly, a GAF is used to encode the one-dimensional vibration signal. The generated timing diagram can maintain the temporal correlation of the signal in the image data without losing the characteristics of the original signal. Then, the ECA module is used to optimize the model, so that the model can adaptively allocate computing resources. Details of the features associated with the type of fault can be captured, and the learning ability of the model is improved. Finally, by comparison with other models, the test proves that the DRCNN not only has a stable structure but also has a low calculation cost, which can meet the potential demand of actual diagnosis.
The remainder of this paper is organized as follows: GAF, depthwise separable convolution, and ECA are briefly reviewed in
Section 2.
Section 3 presents the design of the model structure.
Section 4 tests the presented method and shows the results.
Section 5 concludes this paper.
Author Contributions
Conceptualization, X.G. and Y.X.; methodology, X.G. and T.L.; software, X.G. and Y.X.; validation, X.G. and Y.T.; formal analysis, Y.T.; investigation, Y.T.; resources, Y.T.; data curation, Y.X. and T.L.; writing—original draft preparation, X.G.; writing—review and editing, Y.X. All authors have read and agreed to the published version of the manuscript.
Funding
This work was supported by the Scientific Research Fund of the Department of Education of Liaoning Province, China (LJKZZ20220037).
Data Availability Statement
The data presented in this study are available upon request from the corresponding author. The raw/processed data needed to reproduce these findings cannot be shared publicly at this time, as they are also part of an ongoing study.
Conflicts of Interest
The authors declare no conflict of interest.
References
- Bildirici, M.; Ersin, O. An Investigation of the Relationship between the Biomass Energy Consumption, Economic Growth and Oil Prices. In Proceedings of the 4th International Conference on Leadership, Technology, Innovation and Business Management (ICLTIBM-2014), Istanbul, Turkey, 19–21 November 2014. [Google Scholar]
- Bildirici, M.; Ersin, O. Nexus between Industry 4.0 and environmental sustainability: A Fourier panel bootstrap cointegration and causality analysis. J. Clean. Prod. 2023, 386, 135786. [Google Scholar] [CrossRef]
- Xiao, Q.J.; Luo, Z.H. Early fault diagnosis of the rolling bearing based on the weak signal detection technology. In Proceedings of the 2016 14th International Conference on Control, Automation, Robotics and Vision (ICARCV), Phuket, Thailand, 13–15 November 2016. [Google Scholar]
- Chen, X.F.; Guo, Y.J.; Xu, C.B.; Shang, H.B. Summary of research on fault diagnosis and health monitoring of wind power equipment. China Mech. Eng. 2020, 31, 175–189. [Google Scholar]
- Gao, Z.; Cecati, C.; Ding, S.X. A survey of fault diagnosis and fault-tolerant techniques-Part I: Fault diagnosis with model-based and signal-based approaches. IEEE Trans. Ind. Electron. 2015, 62, 3757–3767. [Google Scholar] [CrossRef]
- Chen, H.M.; Meng, W.; Li, Y.J.; Xiong, Q. An anti-noise fault diagnosis approach for rolling bearings based on multiscale CNN-LSTM and a deep residual learning model. Meas. Sci. Technol. 2023, 34, 0450130. [Google Scholar] [CrossRef]
- Jiang, Z.H.; Zhang, K.; Xiang, L.; Yu, G.; Xu, Y.G. A time-frequency spectral amplitude modulation method and its applications in rolling bearing fault diagnosis. Mech. Syst. Signal Process. 2023, 185, 109832. [Google Scholar] [CrossRef]
- Wang, X.L.; Yan, X.L.; He, Y.L. Fault diagnosis of wind turbine bearing based on IEWT energy order spectrum under variable speed. Acta Energ. Sol. Sin. 2021, 42, 479–486. [Google Scholar]
- Wang, H.Q.; Sun, L.K.; Chen, H.P. Feature extraction method based on pseudo-Wigner-Ville distribution for rotational machinery in variable operating conditions. Chin. J. Mech. Eng. 2011, 24, 661–668. [Google Scholar] [CrossRef]
- Tan, Y.; Sun, W.L.; Wen, G.R.; Huang, X. Wind turbine bearing deterioration trend identification and fault diagnosis based on EWT-MDS. Acta Energ. Sol. Sin. 2018, 39, 3511–3518. [Google Scholar]
- Chen, Z.; Gryllias, K.; Li, W. Mechanical fault diagnosis using convolutional neural networks and extreme learning machine. Mech. Syst. Signal Process. 2019, 133, 106272. [Google Scholar] [CrossRef]
- Pham, M.T.; Kim, J.M.; Kim, C.H. Deep learning-based bearing fault diagnosis method for embedded systems. Sensors 2020, 20, 6886. [Google Scholar] [CrossRef] [PubMed]
- Wang, N.N.; Ma, P.; Zhang, H.L.; Wang, C. Fault diagnosis of rolling bearing based on feature fusion of multi-scale deep convolutional network. Acta Energ. Sol. Sin. 2022, 43, 351–358. [Google Scholar]
- He, K.M.; Zhang, X.Y.; Ren, S.Q.; Sun, J. Deep Residual Learning for Image Recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
- Sandler, M.; Howard, A.; Zhu, M.L.; Zhmoginov, A.; Chen, L.C. Mobilenetv2: Inverted Residuals and Linear Bottlenecks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–23 June 2018; pp. 4510–4520. [Google Scholar]
- Howard, A.G.; Zhu, M.; Chen, B.; Kalenichenko, D.; Wang, W.; Weyand, T.; Andreetto, M.; Adam, H. Mobilenets: Efficient Convolutional Neural Networks for Mobile Vision Applications. arXiv 2017, arXiv:1704.04861. [Google Scholar]
- Wang, Z.; Oates, T. Encoding Time Series as Images for Visual Inspection and Classification Using Tiled Convolutional Neural Networks. In Proceedings of the Workshops at the Twenty-Ninth AAAI Conference on Artificial Intelligence, Boston, MA, USA, 25–29 January 2015. [Google Scholar]
- Wang, Q.L.; Wu, B.G.; Zhu, P.F.; Li, P.H.; Hu, Q.H. Supplementary material for ‘ECA-Net: Efficient channel attention for deep convolutional neural networks. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 13–19 June 2020; pp. 13–19. [Google Scholar]
- Ioffe, S.; Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In Proceedings of the 32nd International Conference on Machine Learning (ICML 2015), Lille, France, 6–11 July 2015; pp. 448–456. [Google Scholar]
- Howard, A.; Sandler, M.; Chu, G.; Chen, L.C.; Chen, B.; Tan, M.; Wang, W.; Zhu, Y.; Pang, R.; Vasudevan, V. Searching for mobilenetv3. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Republic of Korea, 2 October 2019; pp. 1314–1324. [Google Scholar]
| Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2023 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/).