EUVIMEDSkopri EUVIMED
Open access · CC BY 4.0

Robustness of Single-Machine Acoustic Anomaly Detectors to Overlapping Sound: A Multi-Channel Study

Osaowa E. Ogbebor, Jerome Vande Velde, Bruno da Silva

Electronics · 2026-11-20 · 15 (22) · 5413

DOI: 10.3390/electronics15225413 ↗

At a glance

Journal article in Electronics, published online on 2026-11-20. Written by Osaowa E. Ogbebor and 2 co-authors. 43 references. The full text is freely available (open access, CC BY 4.0).

AstrattDettalji biblijografiċiCiteRelated studies

Abstract (original language)

Acoustic anomaly detectors are conventionally trained and tested on a single machine recorded in isolation, yet in industrial environments, multiple machines often operate simultaneously, producing overlapping acoustic signals. This study quantifies the effect of such overlap on detectors trained under isolated-machine conditions. Two unsupervised multi-channel autoencoders, one convolutional and one dense, are trained on normal sound only and reach a mean area under the receiver operating characteristic curve above 0.95 in isolation across four MIMII machine types. When sounds from a second machine, unseen during training, are introduced without retraining, detection performance deteriorates substantially, with considerable variation across conditions. The magnitude of this degradation depends on the target and interfering machine types, as well as on the individual machine units, and abnormalities in the interfering machine frequently trigger false alarms for the target machine. A threshold analysis shows that for some machines the loss is mainly a matter of calibration that re-tuning can recover, whereas for the most transient machine that has intermittent and impulsive acoustic signatures, the detector loses the ability to separate normal from abnormal at any threshold. These findings highlight the limitations of deploying isolation-trained reconstruction-based detectors in environments with overlapping machine sounds and provide a baseline for developing overlap-aware anomaly detection methods.

Cite

APA

Ogbebor, O. E., Velde, J. V., & Silva, B. D. (2026). Robustness of Single-Machine Acoustic Anomaly Detectors to Overlapping Sound: A Multi-Channel Study. Electronics, 15(22), 5413. https://doi.org/10.3390/electronics15225413

Vancouver

Ogbebor OE, Velde JV, Silva BD. Robustness of Single-Machine Acoustic Anomaly Detectors to Overlapping Sound: A Multi-Channel Study. Electronics. 2026;15(22):5413. doi:10.3390/electronics15225413

BibTeX
@article{Ogbebor2026,
  author = {Ogbebor, Osaowa E. and Velde, Jerome Vande and Silva, Bruno da},
  title = {Robustness of Single-Machine Acoustic Anomaly Detectors to Overlapping Sound: A Multi-Channel Study},
  journal = {Electronics},
  year = {2026},
  volume = {15},
  number = {22},
  pages = {5413},
  doi = {10.3390/electronics15225413}
}
Start for free →