Resilient Fault Diagnosis of Rotating Machinery Under Noisy and Harmonic Disturbances
IEEE Access · 2026 · 14 · 62188-62209
DOI: 10.1109/access.2026.3686097 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Míthat Önder. Die Arbeit nennt 86 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Detecting single and compound faults in rotating machinery remains a challenging task, particularly under multi-speed operation and in the presence of severe noise and harmonic disturbances commonly encountered in industrial environments. To address this challenge, this study proposes a robust fault diagnosis approach that combines multi-sensor feature fusion (MSFF) with a hybrid one-dimensional convolutional neural network (1D-CNN). Multi-channel vibration signals obtained from eight sensors at different rotational speeds are reorganized into seven classes, consisting of one healthy class, three single-fault classes, and three compound fault classes, enabling the model to learn speed-invariant fault characteristics. To assess performance under realistic conditions, the model is validated across five datasets, including clean signals, additive white Gaussian noise at different signal-to-noise ratio (SNR) levels, multi-order harmonic disturbances, and their combined effects. In this framework, statistically stable features extracted from both time and frequency domains are integrated directly into the learning process, allowing a more effective interaction between domain knowledge and learned representations. Simulation results show that, while conventional machine learning and deep learning models exhibit noticeable performance degradation under increasingly complex disturbance conditions, the proposed approach maintains stable and high diagnostic performance across all scenarios. The model achieves an overall accuracy of $99.92~\pm ~0.07$ % under clean conditions and maintains strong performance, with an accuracy of $91.62~\pm ~0.89$ %, even under the most challenging composite disturbance case. Further analyses, including statistical significance tests, ablation studies, and computational evaluations, confirm that these improvements are consistent and achieved without introducing excessive computational cost. These results indicate that the proposed hybrid 1D-CNN provides a reliable and efficient solution for fault diagnosis in realistic industrial environments.
Zitieren
APA
Önder, M. (2026). Resilient Fault Diagnosis of Rotating Machinery Under Noisy and Harmonic Disturbances. IEEE Access, 14, 62188-62209. https://doi.org/10.1109/access.2026.3686097
Vancouver
Önder M. Resilient Fault Diagnosis of Rotating Machinery Under Noisy and Harmonic Disturbances. IEEE Access. 2026;14:62188-62209. doi:10.1109/access.2026.3686097
BibTeX
@article{nder2026,
author = {Önder, Míthat},
title = {Resilient Fault Diagnosis of Rotating Machinery Under Noisy and Harmonic Disturbances},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {62188--62209},
doi = {10.1109/access.2026.3686097}
}
EUVIMED