University of St. Gallen

← Publications

From Noise to Knowledge: A Comparative Study of Acoustic Anomaly Detection Models in Pumped-storage Hydropower Plants

Karim Khamaisi, Nicolas Keller, Stefan Krummenacher, Valentin Huber, Bernhard Fässler, Bruno Rodrigues

Workshop on Longevity in IoT Systems (LongevIoT), ACM International Conference on the Internet of Things (IoT) (venue) · Vienna, Austria, 18-21 November 2025 · 2025 · Best Paper Award

Preprint DOI
Figure from From Noise to Knowledge: A Comparative Study of Acoustic Anomaly Detection Models in Pumped-storage Hydropower Plants

Hearing a fault inside a hydropower machine is easy in theory and hard in practice: the environment is deafeningly noisy and clean examples of failures are scarce. This study benchmarks acoustic anomaly detection models on real pumped-storage hydropower recordings, mapping out what actually works when labeled data is limited and conditions are far from lab-clean. It won the Best Paper Award at the LongevIoT workshop of ACM IoT 2025.

Read the full paper (PDF) →

Cite this paper (BibTeX)
@inproceedings{khamaisi2025-noise-to-knowledge,
  author    = {Karim Khamaisi and Nicolas Keller and Stefan Krummenacher and Valentin Huber and Bernhard Fässler and Bruno Rodrigues},
  title     = {{From Noise to Knowledge: A Comparative Study of Acoustic Anomaly Detection Models in Pumped-storage Hydropower Plants}},
  booktitle = {Workshop on Longevity in IoT Systems (LongevIoT), ACM International Conference on the Internet of Things (IoT)},
  year      = {2025},
  address   = {Vienna, Austria},
  url       = {https://arxiv.org/pdf/2509.22881}
}