Synthetic data generation of vibration signals at different speed and load conditions of transmissions utilizing generative adversarial networks
-
Timo König
,
Fabian Wagner
Abstract
Condition monitoring of machines and powertrain components is an essential part of ensuring reliability and product safety in many industries. The monitored machines and components are often divided into different condition classes as well as classified using machine learning methods. In order to enable classification with machine learning algorithms, the acquisition of a sufficient amount of data from each condition class is essential. In reality, the collection of data for faulty system states turns out to be much more difficult, therefore in many use cases balanced data sets are not available. However, when classifying faulty states, an identical number of data per class is of great importance. This problem can be counteracted with synthetic data generation. Generative Adversarial Networks (GAN) are a suitable approach to generate synthetic data based on real measured data. In most cases of synthetic data generation, different damage cases, e.g. from a transmission, are simulated, but a generation of synthetic data is not performed at different operating conditions. However, different speeds and torques are a reality when monitoring, as the drive systems operate under changing operating conditions. Therefore, in the context of this paper, synthetic data generation at different operating states is investigated in order to implement a condition monitoring system for good and bad system conditions which includes different operating states. So, vibration data is acquired at different operating conditions of a transmission on a drive test rig and relevant features are highlighted using a suitable signal pre-processing method. The features, caused by different operating conditions, can also be generated synthetically by GAN. Therefore, it is possible to achieve a similar classification accuracy by integrating synthetically generated data as with real data, which makes the synthetic data generation a viable solution for extending existing data sets.
Zusammenfassung
Die Zustandsüberwachung von Maschinen und Komponenten des Antriebsstrangs ist in vielen Industriezweigen ein wesentlicher Bestandteil zur Gewährleistung der Zuverlässigkeit und Produktsicherheit. Die überwachten Maschinen und Komponenten werden häufig in verschiedene Zustandsklassen eingeteilt und mit Methoden des Machine Learning klassifiziert. Um eine Klassifizierung mit Algorithmen des Machine Learning zu ermöglichen, ist die Erfassung einer ausreichenden Menge an Daten aus jeder Zustandsklasse unerlässlich. In der Realität gestaltet sich die Erfassung von Daten für fehlerhafte Systemzustände deutlich schwieriger, sodass in vielen Anwendungsfällen keine ausgewogenen Datensätze zur Verfügung stehen. Bei der Klassifizierung von fehlerhaften Zuständen ist jedoch eine identische Anzahl an Daten pro Klasse von großer Bedeutung. Diesem Problem kann mit der Generierung künstlicher Daten entgegengewirkt werden. Generative Adversarial Networks (GAN) sind ein geeigneter Ansatz, um künstliche Daten auf Basis realer Messdaten zu erzeugen. In den meisten Fällen der künstlichen Datengenerierung werden zwar verschiedene Schadensfälle z.B. aus einem Getriebe simuliert, eine Generierung künstlicher Daten bei verschiedenen Betriebszuständen wird jedoch nicht durchgeführt. Unterschiedliche Drehzahlen und Drehmomente sind jedoch Realität bei der Überwachung, da die Antriebssysteme unter wechselnden Betriebsbedingungen arbeiten. Daher wird im Rahmen dieser Arbeit die Erzeugung künstlicher Daten bei verschiedenen Betriebszuständen untersucht, um ein Zustandsüberwachungssystem für gute und schlechte Systemzustände zu implementieren, das verschiedene Betriebszustände umfasst. Dazu werden Schwingungsdaten bei unterschiedlichen Betriebszuständen eines Getriebes auf einem Antriebsprüfstand erfasst und relevante Merkmale durch eine geeignete Signalvorverarbeitung hervorgehoben. Die Merkmale, die durch unterschiedliche Betriebszustände verursacht werden, können auch künstlich durch GAN erzeugt werden. Durch die Integration künstlich erzeugter Daten kann daher eine ähnliche Klassifikationsgenauigkeit erreicht werden wie mit realen Daten, was die künstliche Datengenerierung zu einer praktikablen Lösung zur Erweiterung bestehender Datensätze macht.
Funding source: Deutsche Forschungsgemeinschaft
Award Identifier / Grant number: INST 52/19-1 FUGB
Funding source: European Union, the state of Baden-Württemberg, the district of Ostalb and the city of Aalen
Award Identifier / Grant number: FEIH_2486678
About the authors
Timo König is a research assistant at the Institute of Driveline Technology Aalen (IAA) at Aalen University. He previously completed his Bachelor’s and Master’s degree in Mechanical Engineering at Baden-Wuerttemberg Cooperative State University Heidenheim and Aalen University. His research focuses on condition monitoring of machine components and synthetic data generation.
Fabian Wagner is a research assistant at the Faculty for Electronics and Computer Science at Aalen University. He previously completed his Research Master’s degree in Mechanical Engineering at Aalen University. His research focuses on the condition monitoring of machine components.
Robin Bäßler studied mechanical engineering at Aalen University. After completing his research master’s degree in Advanced Materials and Manufacturing, he started working as an academic assistant at the Institute of Driveline Technology Aalen (IAA).
Prof. Dr.-Ing. Markus Kley is Professor for Design Theory and Drive Technology at the IAA – Institute for Drive Technology at Aalen University. He studied mechanical engineering at the University of Stuttgart and engineering and mechanics at the Georgia Institute of Technology, Atlanta, USA. His doctoral thesis was on the influence of transmission damage in city buses. He is responsible for several test benches at the IAA. His research interests focus on the development and testing of future powertrain technologies.
Marcus Liebschner received the Dipl.-Ing. (FH) and M. Eng. in electrical engineering from University of Applied Sciences Aalen and from University of Applied Sciences Würzburg-Schweinfurt. He received his Dr.-Ing. degree in electrical engineering from the Technical University Ilmenau in 2009. He worked also for the Voith Hydro company. Since 2012, he is working as a Professor for electrical engineering at the University of Applied Sciences in Aalen.
-
Author contributions: All the authors have accepted responsibility for the entire content of this submitted manuscript and approved submission.
-
Research funding: This work was funded by the European Union, the state of Baden-Württemberg, the district of Ostalb and the city of Aalen as part of the project “KI-Werkstatt Mittelstand” under the funding code FEIH_2486678. The tests were carried out with the help of the road-to-rig vehicle test bench “VAPS”, which was funded by the German Research Foundation (DFG funding code: INST 52/19-1 FUGB).
-
Declaration of conflicts: The authors declare no conflicts of interests in preparing this article.
References
[1] Schaeffler Technologies, Condition Monitoring Praxis. Handbuch zur Schwingungs-Zustandsüberwachung von Maschinen und Anlagen, Mainz am Rhein, Vereinigte Fachverlage, 2019.Search in Google Scholar
[2] Schaeffler Technologies, Wälzlagerpraxis. Handbuch zur Gestaltung und Berechnung von Wälzlagerungen, Mainz am Rhein, Vereinigte Fachverlage, 2019.Search in Google Scholar
[3] T. König, R. Bader, and M. Kley, “Schwingungsbasierte Fehlererkennung und Schadensdetektion an Getrieben durch Einbindung von Methoden des Machine Learning. 3. VDI-Fachtagung Schwingungen 2021,” VDI-Ber., vol. 2391, pp. 53–66, 2021. https://doi.org/10.51202/9783181023914-53.Search in Google Scholar
[4] R. B. Randall, Vibration-based Condition Monitoring. Industrial, Automotive and Aerospace Applications, Hoboken, NJ, Wiley, 2022.10.1002/9781119477631Search in Google Scholar
[5] F. Wagner, T. König, M. Benninger, M. Kley, and M. Liebschner, “Generation of synthetic data with low-dimensional features for condition monitoring utilizing Generative Adversarial Networks,” Procedia Comput. Sci., vol. 207, no. 4, pp. 634–643, 2022. https://doi.org/10.1016/j.procs.2022.09.118.Search in Google Scholar
[6] T. König, M. Kley, F. Wagner, and M. Liebschner, “Enhanced damage classification on transmissions by generating synthetic data with Generative Adversarial Networks (GAN),” in International Conference on Gears 2022, VDI Verlag, 2022, pp. 227–238.10.51202/9783181023891-227Search in Google Scholar
[7] M. Bauer, F. Wagner, and M. Kley, “Optimierung der Sensorpositionierung bei schwingungsbasierter Wälzlagerzustandsüberwachung unter Einbezug von Systemeigenmoden,” TM – Tech. Mess., vol. 88, no. 11, pp. 674–685, 2021. https://doi.org/10.1515/teme-2021-0045.Search in Google Scholar
[8] M. Bauer, D. Proksch, J. Kopetschek, F. Wagner, and M. Kley, “Entwicklung und Validierung einer Methode zur Ermittlung der minimalen Performanceanforderungen an Sensoren für die schwingungsbasierte Zustandsüberwachung. 3. VDI-Fachtagung Schwingungen 2021,” VDI-Ber., vol. 2391, pp. 89–104, 2021. https://doi.org/10.51202/9783181023914-89.Search in Google Scholar
[9] M. Bauer, N. Balaratnam, J. Weidenauer, F. Wagner, and M. Kley, “Comparison of envelope demodulation methods in the analysis of rolling bearing damage,” J. Vib. Control, 2022, Art. no. 107754632211291, https://doi.org/10.1177/10775463221129155.Search in Google Scholar
[10] T. König, R. Bader, S. Pandit, and M. Kley, “Getriebespezifische Schadensanalyse an elektromechanischen Antriebssystemen unter Verwendung mehrerer Beschleunigungssensoren und künstlich neuronaler Netze. 8,” in IFToMM D-A-CH Konferenz, 2022, 24./25. Februar 2022.Search in Google Scholar
[11] T. Zhang, J. Chen, F. Li, et al.., “Intelligent fault diagnosis of machines with small & imbalanced data: a state-of-the-art review and possible extensions,” ISA Trans., vol. 119, no. 2022, pp. 152–171, 2022. https://doi.org/10.1016/j.isatra.2021.02.042.Search in Google Scholar PubMed
[12] W. Mao, Y. Liu, L. Ding, and Y. Li, “Imbalanced Fault diagnosis of rolling bearing based on generative adversarial network: a comparative study,” IEEE Access, vol. 7, pp. 9515–9530, 2019. https://doi.org/10.1109/ACCESS.2018.2890693.Search in Google Scholar
[13] M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein generative adversarial networks,” in International Conference on Machine Learning, 2017, pp. 214–223.Search in Google Scholar
[14] G. Yang, Y. Zhong, L. Yang, H. Tao, J. Li, and R. Du, “Fault diagnosis of harmonic drive with imbalanced data using generative adversarial network,” IEEE Trans. Instrum. Meas., vol. 70, pp. 1–11, 2021. https://doi.org/10.1109/TIM.2021.3089240.Search in Google Scholar
[15] S. Fahle, T. Glaser, A. Kneißler, and B. Kuhlenkötter, “Improving quality prediction in radial-axial ring rolling using a semi-supervised approach and generative adversarial networks for synthetic data generation,” Prod. Eng., vol. 16, no. 1, pp. 175–185, 2022. https://doi.org/10.1007/s11740-021-01075-x.Search in Google Scholar
[16] M. Zheng, T. Li, R. Zhu, et al.., “Conditional Wasserstein generative adversarial network-gradient penalty-based approach to alleviating imbalanced data classification,” Inf. Sci., vol. 512, pp. 1009–1023, 2020. https://doi.org/10.1016/j.ins.2019.10.014.Search in Google Scholar
[17] H. Zhang, R. Wang, R. Pan, and H. Pan, “Imbalanced Fault diagnosis of rolling bearing using enhanced generative adversarial networks,” IEEE Access, vol. 8, pp. 185950–185963, 2020. https://doi.org/10.1109/ACCESS.2020.3030058.Search in Google Scholar
[18] X. Xiong, J. Hongkai, X. Li, and M. Niu, “A Wasserstein gradient-penalty generative adversarial network with deep auto-encoder for bearing intelligent fault diagnosis,” Meas. Sci. Technol., vol. 31, p. 4, 2020. https://doi.org/10.1088/1361-6501/ab47df.Search in Google Scholar
[19] J. Zhao and W. Huang, “Transfer learning method for rolling bearing fault diagnosis under different working conditions based on CycleGAN,” Meas. Sci. Technol., vol. 33, no. 2, p. 25003, 2022. https://doi.org/10.1088/1361-6501/ac3942.Search in Google Scholar
[20] F. Naaz, A. Herle, J. Channegowda, A. Raj, and M. Lakshminarayanan, “A generative adversarial network-based synthetic data augmentation technique for battery condition evaluation,” Int. J. Energy Res., vol. 45, no. 13, pp. 19120–19135, 2021. https://doi.org/10.1002/er.7013.Search in Google Scholar
[21] Z. Niu, M. Z. Reformat, W. Tang, and B. Zhao, “Electrical equipment identification method with synthetic data using edge-oriented generative adversarial network,” IEEE Access, vol. 8, pp. 136487–136497, 2020. https://doi.org/10.1109/ACCESS.2020.3011689.Search in Google Scholar
[22] G. Zhang, H. Xiao, J. Jiang, Q. Liu, Y. Liu, and L. Wang, “A multi-index generative adversarial network for tool wear detection with imbalanced data,” Complexity, vol. 2020, no. 4, pp. 1–10, 2020. https://doi.org/10.1155/2020/5831632.Search in Google Scholar
[23] Q.-X. Zhu, T. Xu, Y. Xu, and Y.-L. He, “Improved virtual sample generation method using enhanced conditional generative adversarial networks with cycle structures for soft sensors with limited data,” Ind. Eng. Chem. Res., vol. 61, no. 1, pp. 530–540, 2022. https://doi.org/10.1021/acs.iecr.1c03197.Search in Google Scholar
[24] R. Bäßler, T. Bäßler, and M. Kley, “Classification of load and rotational speed at wire-race bearings using convolutional neural networks with vibration spectrogram,” TM – Tech. Mess., vol. 89, no. 5, pp. 352–362, 2022. https://doi.org/10.1515/teme-2021-0143.Search in Google Scholar
[25] T. Bäßler, R. Bäßler, and M. Kley, “Augmented mel-spectrogram VGG-16 model for axial and radial load classification at wire-race bearings,” TM – Tech. Mess., vol. 89, no. 9, pp. 573–579, 2022. https://doi.org/10.1515/teme-2022-0039.Search in Google Scholar
© 2023 Walter de Gruyter GmbH, Berlin/Boston
Articles in the same Issue
- Frontmatter
- Research Articles
- Experimental investigation of liquid metal droplet flow affected by a time-dependent magnetic field
- Synthetic data generation of vibration signals at different speed and load conditions of transmissions utilizing generative adversarial networks
- Quantifizierung der Klassifikationsleistung von Oberflächeninspektionssystemen in der Flachstahlproduktion
- A reinforced corrosion assessment method based on a new magnetic sensor and improved adaptive filtering
- Estimators of tissue absorption parameters power-law prefactor and power-law exponent from medical ultrasonic images
- Linearity test of triangular waveform generators
Articles in the same Issue
- Frontmatter
- Research Articles
- Experimental investigation of liquid metal droplet flow affected by a time-dependent magnetic field
- Synthetic data generation of vibration signals at different speed and load conditions of transmissions utilizing generative adversarial networks
- Quantifizierung der Klassifikationsleistung von Oberflächeninspektionssystemen in der Flachstahlproduktion
- A reinforced corrosion assessment method based on a new magnetic sensor and improved adaptive filtering
- Estimators of tissue absorption parameters power-law prefactor and power-law exponent from medical ultrasonic images
- Linearity test of triangular waveform generators
