Enhancing the generalizability of public building energy system fault detection method: A research on unknown multi-source fault detection and diagnosis method based on data-driven heuristic reasoning (DHR)

Zhang, Boyan, Wang, Jiaming, Rezgui, Yacine ORCID: https://orcid.org/0000-0002-5711-8400 and Zhao, Tianyi 2025. Enhancing the generalizability of public building energy system fault detection method: A research on unknown multi-source fault detection and diagnosis method based on data-driven heuristic reasoning (DHR). Energy , 137841. 10.1016/j.energy.2025.137841

Full text not available from this repository.

Abstract

The reliable operation of the building energy system (BES) is the basis for the efficient use of energy in buildings. To ensure that BES is regulated in real-time through control strategies to improve building energy efficiency and indoor environmental comfort, numerous sensors are usually installed. However, faulty sensors data pose a great challenge to the healthy operation of the control loop of the BES. In addition, conventional BES fault diagnosis and detection (FDD) methods rely heavily on the abundance of historical sample data. Unfortunately, especially multi-source faults data, obtaining the full variety and abundance of fault samples to construct FDD models is extremely difficult in BES. To address this challenging, it proposes a data-driven heuristic reasoning (DHR)-based fault diagnosis model in this paper. The proposed method relies on only partial fault type samples as known fault data for FDD model training to achieve multi-source fault diagnosis for both known and unknown types in BES. It discusses in three cases, each of which discusses a potential nine fault conditions depending on the number of unknown faults. In the experiments with 27 fault conditions, the proposed method achieves a diagnosis rate of 99.98-100% for known multi-source faults and 87.1-100% for unknown faults. The research results greatly improve the generality of current BES fault diagnosis methods and reduce the dependence of FDD methods on training sample data.

Item Type:	Article
Date Type:	Published Online
Status:	In Press
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	0360-5442
Date of Acceptance:	30 July 2025
Last Modified:	12 Aug 2025 08:30
URI:	https://orca.cardiff.ac.uk/id/eprint/180345

Actions (repository staff only)

Edit Item