LSTM Autoencoder-Based Real-Time Anomaly Detection for Univariate Hydraulic Turbine Time-Series Data
IEEE Sensors Journal · 15.07.2026 · 26 (14) · 21666-21679
DOI: 10.1109/jsen.2026.3698747 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 15.07.2026. Verfasst von Blondelle Melina Atsafack und 3 weiteren Autor:innen. Die Arbeit nennt 65 Literaturangaben. Gefördert von: International Centre of Insect Physiology and Ecology.
Abstract (Originalsprache)
Hydraulic turbines are essential components for energy generation, and their reliable operation is crucial for minimizing downtime and maintenance costs. This study introduces a real-time health monitoring system for hydraulic turbines, utilizing long short-term memory autoencoders (LSTM-AEs) to detect anomalies in vibration time-series data. To reduce the incidence of false alarms, both supervised and unsupervised learning methods were examined. The supervised LSTM-AE, trained on historical vibration data, demonstrated superior performance with an accuracy of 100% and a precision of 99% in identifying abnormal conditions. In contrast, the unsupervised LSTM-AE achieved an accuracy of 90% and a precision of 65%, indicating a lower discriminatory capability. The selected model was integrated into the LabVIEW platform to facilitate real-time anomaly detection, providing continuous monitoring through an intuitive graphical interface. A data point is classified as abnormal if its reconstruction error surpasses a predefined threshold, and an alert is generated if more than half of the samples within a sequence are identified as abnormal. These results underscore the effectiveness of supervised learning in industrial anomaly detection and support predictive maintenance (PdM) strategies that enhance the reliability and sustainability of hydropower systems.
Förderung
- International Centre of Insect Physiology and Ecology
Zitieren
APA
Atsafack, B. M., Juma, D., Rushingabigwi, G., & Kabiri, C. (2026). LSTM Autoencoder-Based Real-Time Anomaly Detection for Univariate Hydraulic Turbine Time-Series Data. IEEE Sensors Journal, 26(14), 21666-21679. https://doi.org/10.1109/jsen.2026.3698747
Vancouver
Atsafack BM, Juma D, Rushingabigwi G, Kabiri C. LSTM Autoencoder-Based Real-Time Anomaly Detection for Univariate Hydraulic Turbine Time-Series Data. IEEE Sensors Journal. 2026;26(14):21666-21679. doi:10.1109/jsen.2026.3698747
BibTeX
@article{Atsafack2026,
author = {Atsafack, Blondelle Melina and Juma, Denis and Rushingabigwi, Gerard and Kabiri, Charles},
title = {LSTM Autoencoder-Based Real-Time Anomaly Detection for Univariate Hydraulic Turbine Time-Series Data},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {14},
pages = {21666--21679},
doi = {10.1109/jsen.2026.3698747}
}
EUVIMED