Details
Original language | English |
---|---|
Pages (from-to) | 11305-11310 |
Number of pages | 6 |
Journal | IFAC-PapersOnLine |
Volume | 56 |
Issue number | 2 |
Early online date | 22 Nov 2023 |
Publication status | Published - 2023 |
Abstract
The implementation effort of data-driven fault diagnosis systems greatly exceeds its economic benefits in many industrial cases. Consequently, highly adapted, individual solutions instead of widespread distribution in the market are currently the result. The biggest problem is the availability of large amounts of labeled data, in particular fault data. In this work, we propose a multi-source domain adaptation procedure that integrates synthetic fault data generation into cross-domain classifier training to overcome this issue. The approach does not require fault data in the target domain which is highly relevant in practice. It is examined using the rarely studied example of diagnosing faulty pretensioning of belt drives. Datasets for multiple domains are collected by attaching different loads to the machine. An extensive experimental study on single and multiple source domains demonstrates the effectiveness of the proposed approach. The generation of fault data outperforms the benchmark methods, especially for multi-source scenarios. Overall, the cross-domain fault diagnosis of belt drives yields promising results to enable a broad range of industrial applications.
Keywords
- Domain adaptation, Fault detection and diagnosis, Machine learning, Mechatronic systems, Motion control systems, Time series modeling
ASJC Scopus subject areas
- Engineering(all)
- Control and Systems Engineering
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: IFAC-PapersOnLine, Vol. 56, No. 2, 2023, p. 11305-11310.
Research output: Contribution to journal › Article › Research › peer review
}
TY - JOUR
T1 - Multi-Source Domain Adaptation for Fault Diagnosis of Belt Drives
AU - Fehsenfeld, Moritz
AU - Kühn, Johannes
AU - Kortmann, Karl-Philipp
N1 - Publisher Copyright: Copyright © 2023 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)
PY - 2023
Y1 - 2023
N2 - The implementation effort of data-driven fault diagnosis systems greatly exceeds its economic benefits in many industrial cases. Consequently, highly adapted, individual solutions instead of widespread distribution in the market are currently the result. The biggest problem is the availability of large amounts of labeled data, in particular fault data. In this work, we propose a multi-source domain adaptation procedure that integrates synthetic fault data generation into cross-domain classifier training to overcome this issue. The approach does not require fault data in the target domain which is highly relevant in practice. It is examined using the rarely studied example of diagnosing faulty pretensioning of belt drives. Datasets for multiple domains are collected by attaching different loads to the machine. An extensive experimental study on single and multiple source domains demonstrates the effectiveness of the proposed approach. The generation of fault data outperforms the benchmark methods, especially for multi-source scenarios. Overall, the cross-domain fault diagnosis of belt drives yields promising results to enable a broad range of industrial applications.
AB - The implementation effort of data-driven fault diagnosis systems greatly exceeds its economic benefits in many industrial cases. Consequently, highly adapted, individual solutions instead of widespread distribution in the market are currently the result. The biggest problem is the availability of large amounts of labeled data, in particular fault data. In this work, we propose a multi-source domain adaptation procedure that integrates synthetic fault data generation into cross-domain classifier training to overcome this issue. The approach does not require fault data in the target domain which is highly relevant in practice. It is examined using the rarely studied example of diagnosing faulty pretensioning of belt drives. Datasets for multiple domains are collected by attaching different loads to the machine. An extensive experimental study on single and multiple source domains demonstrates the effectiveness of the proposed approach. The generation of fault data outperforms the benchmark methods, especially for multi-source scenarios. Overall, the cross-domain fault diagnosis of belt drives yields promising results to enable a broad range of industrial applications.
KW - Domain adaptation
KW - Fault detection and diagnosis
KW - Machine learning
KW - Mechatronic systems
KW - Motion control systems
KW - Time series modeling
UR - http://www.scopus.com/inward/record.url?scp=85184960225&partnerID=8YFLogxK
U2 - 10.1016/j.ifacol.2023.10.411
DO - 10.1016/j.ifacol.2023.10.411
M3 - Article
VL - 56
SP - 11305
EP - 11310
JO - IFAC-PapersOnLine
JF - IFAC-PapersOnLine
SN - 2405-8963
IS - 2
ER -