EUVIMEDEUVIMED entdecken
Crossref

Online Industrial Sensor Drift Identification Using Anomaly Detection Methods

Bruno S. Silva, Clayton H. Costa, Bruna M. Mulinari, André E. Lazzaretti

IEEE Sensors Letters · 2026-08 · 10 (8) · 6006804-6006804

DOI: 10.1109/lsens.2026.3707698 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Letters, 2026-08. Verfasst von Bruno S. Silva und 3 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Reliable sensor drift detection is essential in industrial environments to preserve measurement integrity, maintain product quality, and reduce productivity losses caused by undetected instrument deviations. This letter presents an online workflow for industrial sensor drift identification using anomaly-detection models deployed within a production-line monitoring system. The approach is motivated by a real diesel injector test bench in which calibration is conventionally verified every 4 h, allowing drift conditions to affect multiple samples before corrective action is taken. The proposed strategy learns normal measurement behavior using only no-drift data and identifies abnormal conditions as they emerge during operation. Two unsupervised models were evaluated, namely, a variational autoencoder (VAE) and a one-class support vector machine one-class support vector machine (OC-SVM), both combined with a debounce filter to reduce false alarms. SHapley Additive exPlanations-based feature selection was also investigated to reduce input dimensionality and computational cost. Experiments were conducted using eight weeks of industrial data, including 9129 training and 2282 validation samples, and a 401-sample shadow mode evaluation set containing 51 anomalous samples from two drift conditions observed in operation. The VAE outperformed the OC-SVM with the full feature set, while the reduced-feature VAE achieved no classification errors in the evaluated shadow mode dataset and an F1-score of 1.0. Inference times below 7 ms were recorded from the online deployment within the test cycle. Future work includes evaluations with additional drift scenarios.

Zitieren

APA

Silva, B. S., Costa, C. H., Mulinari, B. M., & Lazzaretti, A. E. (2026). Online Industrial Sensor Drift Identification Using Anomaly Detection Methods. IEEE Sensors Letters, 10(8), 6006804-6006804. https://doi.org/10.1109/lsens.2026.3707698

Vancouver

Silva BS, Costa CH, Mulinari BM, Lazzaretti AE. Online Industrial Sensor Drift Identification Using Anomaly Detection Methods. IEEE Sensors Letters. 2026;10(8):6006804-6006804. doi:10.1109/lsens.2026.3707698

BibTeX
@article{Silva2026,
  author = {Silva, Bruno S. and Costa, Clayton H. and Mulinari, Bruna M. and Lazzaretti, André E.},
  title = {Online Industrial Sensor Drift Identification Using Anomaly Detection Methods},
  journal = {IEEE Sensors Letters},
  year = {2026},
  volume = {10},
  number = {8},
  pages = {6006804--6006804},
  doi = {10.1109/lsens.2026.3707698}
}
Kostenlos starten →