Abstract
Wire-race bearings are designed to be space-saving and are used in settings, where high precision under continuous high load and torque is required. Thus, load detection is a vital part of their condition monitoring. Measuring load using load cells defeats the space-saving purpose of these bearings. Using virtual sensors, by measuring axial and radial loading indirectly, via an accelerometer attached to the housing of the bearing, can provide an alternative.
We extend an earlier approach of a virtual sensor using vibration spectrograms with convolutional neural networks. We propose to use transfer learning to reduce overfitting. Moreover we use Mel-spectrograms instead of standard spectrograms and apply feature augmentation techniques. Thereby the virtual sensor improves validation accuracy from 86.3 % to 94.2 %. Using the logarithmic Mel-scale enables to over-sample the characteristic rollover frequencies which appear up to 500 Hz. Frequency masking shows a higher degree of effect compared to normal spectrograms. Further, we find that by using Mel-spectrograms, the combined latency of the spectrogram generation and forward propagation is reduced by 61.6 %.
Zusammenfassung
Drahtwälzlager sind konstruiert um platzsparend in Maschinen eingebaut zu werden in denen eine hohe Präzision bei dauerhaft hoher Belastung und hohem Drehmoment erforderlich ist. Daher ist die Lasterfassung ein wichtiger Bestandteil der Zustandsüberwachung. Die Messung der Belastung mit Hilfe von Kraftmessdosen ist engegen dem platzsparenden Prinzip dieser Lager. Die Verwendung virtueller Sensoren, die die axiale und radiale Belastung indirekt über einen am Lagergehäuse angebrachten Beschleunigungssensor messen, kann eine Alternative darstellen.
Wir erweitern einen früheren Ansatz eines virtuellen Sensors, der Spektrogramme mit Convolutional Neural Networks verwendet. Wir schlagen vor, Transfer Learning einzusetzen, um eine Überanpassung des Modells zu vermeiden. Außerdem verwenden wir Mel-Spektrogramme anstelle von normalen Spektrogrammen und wenden Augmentationstechniken an. Dadurch verbessert der virtuelle Sensor die Validierungsgenauigkeit von 86,3 % auf 94,2 %. Die Verwendung der logarithmischen Mel-Skala ermöglicht eine Überabtastung der charakteristischen Rollover-Frequenzen, die bis zu 500 Hz auftreten. Die Technik der Frequenzmaskierung zeigt im Vergleich zu normalen Spektrogrammen einen höheren Grad an Wirkung. Außerdem stellen wir fest, dass die Verwendung von Mel-Spektrogrammen die kombinierte Latenz der Spektrogramm Erstellung und Vorwärtspropagation des Netzwerkes um 61,1 % reduziert.
About the authors
Tim Bäßler studied economics and political science at the University of Konstanz. He worked as a data analyst for Steinbeis Transfer Platform Industry 4. (TPBW 4.0). His focus was on the implementation of deep learning for condition monitoring. He is currently employed as an internal auditor and data analyst for the Schwarz Group in Neckarsulm.
Robin Bäßler studied 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 drive concepts. There, he is currently working on a research project on the integration of a virtual sensor in a welding process for quality assessment.
Markus Kley studied mechanical engineering and completed his PhD (Dr.-Ing.) at University of Stuttgart. After graduating, he worked in various positions at Voith Turbo in Crailsheim. His last position was head of engine component technology. Subsequently, he was appointed Professor of Design in General Mechanical Engineering at Aalen University. He is coordinator at Aalen University of Applied Sciences of the “Cooperative Doctoral Program PROMISE 4.0” with the participation of Stuttgart University, Esslingen University, Heilbronn University and Aalen University. Further, he is the head for the research focus (System Integration / Optimization - Methodology Testing) in the research building of Aalen University. In addition, he is the head of the Steinbeis Transfer Platform Industry 4.0 (TPBW 4.0) and the Steinbeis Transfer Center for Innovative Drive Technology and Waste Heat Utilization.
References
1. M. Schmidt, H. Spieth, C. Haubach, C. Kühne, Optimisation of material efficiency in the production of wire race bearings. 100 Pioneers in Efficient Resource Management. Springer (2019), pp. 274–277, ISBN: 978-3-662-56744-9.10.1007/978-3-662-56745-6_53Search in Google Scholar
2. J. Aguirrebeitia, M. Inigo, H. Iker, A. Mikel, C. Ibai, Wire twisting stiffness modelling with application in wire race ball bearings. Derivation of analytical formula and finite element validation, Mechanism and Machine Theory (2019), DOI: 10.1016/j.mechmachtheory.2019.05.012.Search in Google Scholar
3. D. Gunia, T. Smolnicki, The influence of the geometrical parameters for stress distribution in wire raceway slewing bearing, Archive of Mechanical Engineering (2017), DOI: 10.1515/meceng-2017-0019.Search in Google Scholar
4. R. Bäßler, T. Bäßler, M. Kley, Classification of load and rotational speed at wire-race bearings using Convolutional Neural Networks with vibration spectrograms, tm-Technisches Messen (2022), DOI: 10.1515/teme-2021-0143.Search in Google Scholar
5. R. Bäßler, M. Bauer, M. Kley, Einsatz eines virtuellen Sensors an einem Drahtwälzlager, Aalener Kolloquium antriebstechnische Anwendungen, AKAA (2020), 112–117, ISBN: 978-3-8440-7529-8.Search in Google Scholar
6. M. Bauer, R. Bäßler, M. Kley, Lastdetektierung durch intelligente Schwingungsanalyse in Leichtbaulagern, Stuttgarter Symposium für Produktentwicklung, SSP (2019), 73–82, DOI: 10.18419/opus-10394.Search in Google Scholar
7. L. Hafner, R. Bäßler, S. Schwarzer, F. Dohnal, M. Kley, Bubble Size Detection by Process Ancillaries Chemie Ingenieur Technik, (2022), DOI: 10.1002/cite.202100175.10.1002/cite.202100175Search in Google Scholar
8. M. Ping, Z. Hongli, F. Wenhui, W. Cong, W. Guangrui, Z. Xining, A novel bearing fault diagnosis method based on 2D image representation and transfer learning-convolutional neural network, Measurement Science and Technology (2019), DOI: 10.1088/1361-6501/ab0793.10.1088/1361-6501/ab0793Search in Google Scholar
9. Z. Wang, Q. Liu, H. Chen, X. Chu, A deformable CNN-DLSTM based transfer learning method for fault diagnosis of rolling bearing under multiple working conditions, International Journal of Production Research (2021), DOI: 10.1080/00207543.2020.1808261.10.1080/00207543.2020.1808261Search in Google Scholar
10. M. Hasan, M. Islam, J. Kim, Acoustic spectral imaging and transfer learning for reliable bearing fault diagnosis under variable speed conditions, Measurement (2019), DOI: 10.1016/j.measurement.2019.02.075.10.1016/j.measurement.2019.02.075Search in Google Scholar
11. T. Lu, F. Yu, B. Han, J. Wang, A generic intelligent bearing fault diagnosis system using convolutional neural networks with transfer learning, IEEE Access (2020), DOI: 10.1109/ACCESS.2020.3022840.10.1109/ACCESS.2020.3022840Search in Google Scholar
12. E. Brusa, C. Delprete, L. Di Maggio, Deep Transfer Learning for Machine Diagnosis: From Sound and Music Recognition to Bearing Fault Detection, Applied Sciences (2021), DOI: 10.3390/app112411663.10.3390/app112411663Search in Google Scholar
13. K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition, arXiv (2014), DOI: 10.48550/arXiv.1409.1556.Search in Google Scholar
14. N. Polyzotis, M. Zaharia, What can Data-Centric AI Learn from Data and ML Engineering?, arXiv (2021), DOI: 10.48550/arXiv.2112.06439.Search in Google Scholar
15. M. Dörfler, R. Bammer, T. Grill, Inside the spectrogram: Convolutional Neural Networks in audio processing, 2017 International Conference on Sampling Theory and Applications (SampTA) (2017), IEEE, DOI: 10.1109/SAMPTA.2017.8024472.10.1109/SAMPTA.2017.8024472Search in Google Scholar
16. A. Howard, Some improvements on deep convolutional neural network based image classification, arXiv (2013), DOI: 10.48550/arXiv.1312.5402.Search in Google Scholar
17. D. Park, W. Chan, Y. Zhang, C. Chiu, B. Zoph, E. Cubuk, Q. Le, Specaugment: A simple data augmentation method for automatic speech recognition, arXiv (2019), DOI: 10.21437/Interspeech.2019-2680.10.21437/Interspeech.2019-2680Search in Google Scholar
18. S. Umesh, L. Cohen, D. Nelson, Fitting the mel scale, 1999 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings. ICASSP99, (1999), IEEE, DOI: 10.1109/ICASSP.1999.758101.10.1109/ICASSP.1999.758101Search in Google Scholar
© 2022 Walter de Gruyter GmbH, Berlin/Boston
