BRICK-Automated Virtual Temperature Sensors for Sensor Fault Detection, Isolation, and Discrimination in Smart-Building HVAC Systems
Sensors · 31.05.2026 · 26 (11) · 3465
DOI: 10.3390/s26113465 ↗Kurz gefasst
Fachartikel in der Zeitschrift Sensors, online erschienen am 31.05.2026. Verfasst von Khaled Chahine und 1 weiteren Autor:innen. Beteiligte Einrichtung: College of Engineering and Technology, American University of the Middle East, Egaila 54200, Kuwait. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Sensor bias faults in closed-loop HVAC systems pose a detection challenge that is both subtle and costly. Because the control loop compensates for biased readings by driving the affected sensor back toward its setpoint, the fault becomes invisible to conventional threshold monitors. The anomaly does not vanish, however; it is redistributed across correlated sensors, disrupting their mutual consistency. We propose a framework that automatically derives virtual temperature sensor models from BRICK schema metadata. LightGBM regressors, trained on fault-free inter-sensor relationships, produce z-scored prediction residuals that serve as detection signals. Fault isolation is achieved by ranking sensors by their median daily anomaly scores; fault-type discrimination relies on analysis of actuator command-position discrepancies. On the Lawrence Berkeley National Laboratory (LBNL) fault detection and diagnosis (FDD) benchmark, the method achieves an area under the receiver operating characteristic curve (AUC) of 0.9992 for the mildest sensor bias (SA +2 °C), an AUC of 1.0 for all other single-duct air handling unit (SD-AHU) scenarios, and an AUC of 1.0 for all fan coil unit (FCU) sensor bias scenarios. In all four SD-AHU sensor bias scenarios, the biased sensor (SA_TEMP) ranks first or second; for the larger biases (±4 °C), SA_TEMP consistently ranks first. A robustness analysis over 10 random seeds confirms that detection AUC remains above 0.997 in all cases. Sensor and mechanical faults fall into non-overlapping clusters in the command-position discrepancy space. On the FCU system, the proposed method substantially outperforms principal component analysis (PCA) (AUC = 1.0 versus 0.63–0.90) and provides diagnostic capabilities not available with PCA. Notably, a single pipeline function handles both system types without modification, confirming cross-system scalability through the BRICK metadata layer. The results confirm that BRICK-automated virtual sensor construction is a viable approach for scalable, deployment-ready sensor validation in smart-building HVAC systems.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- College of Engineering and Technology, American University of the Middle East, Egaila 54200, Kuwait.
- Electrical and Computer Engineering Department, American University of Beirut, Beirut 1107 2020, Lebanon.
- Institut FEMTO-ST, CNRS, IUT-NFC, Université Marie et Louis Pasteur, F-90000 Belfort, France.
ORCID-Profile: Khaled Chahine, Hassan N Noura
Zitieren
APA
Chahine, K., & Noura, H. N. (2026). BRICK-Automated Virtual Temperature Sensors for Sensor Fault Detection, Isolation, and Discrimination in Smart-Building HVAC Systems. Sensors, 26(11), 3465. https://doi.org/10.3390/s26113465
Vancouver
Chahine K, Noura HN. BRICK-Automated Virtual Temperature Sensors for Sensor Fault Detection, Isolation, and Discrimination in Smart-Building HVAC Systems. Sensors. 2026;26(11):3465. doi:10.3390/s26113465
BibTeX
@article{Chahine2026,
author = {Chahine, Khaled and Noura, Hassan N},
title = {BRICK-Automated Virtual Temperature Sensors for Sensor Fault Detection, Isolation, and Discrimination in Smart-Building HVAC Systems},
journal = {Sensors},
year = {2026},
volume = {26},
number = {11},
pages = {3465},
doi = {10.3390/s26113465}
}
EUVIMED