Residual-corrected physics-based modelling for SF6 density prediction and anomaly detection

Tools

Brash, Alistair and Shi, Guo and Pestana, Jennifer and Brown, Blair and Tachtatzis, Christos and Liu, Ting and Irwin, Fiona and Lumsden, Nicol and Stephen, Bruce (2026) Residual-corrected physics-based modelling for SF6 density prediction and anomaly detection. IEEE Transactions on Dielectrics and Electrical Insulation. ISSN 1070-9878 (In Press) (https://doi.org/10.1109/TDEI.2026.3736333)

[thumbnail of Brash-etal-IEEE-TDEI-2026-Residual-corrected-physics-based-modelling-for-SF6-density-prediction-and] Text. Filename: Brash-etal-IEEE-TDEI-2026-Residual-corrected-physics-based-modelling-for-SF6-density-prediction-and.pdf
Accepted Author Manuscript
Restricted to Repository staff only until 1 January 2099.
Download (858kB) | Request a copy

Abstract

Sulphur hexafluoride (SF6) is widely used as an insulating gas in high-voltage electrical equipment due to its excellent electrical insulating properties, including high dielectric strength and thermal stability. These characteristics enable the development of compact and reliable gas-insulated systems. However, although SF6-insulated assets are designed to contain the gas, the containment barriers may, over time, degrade; this creates operational and environmental challenges. Monitoring of SF6 density provides a potential method of detecting and managing such leakages and other abnormal events. Existing alarm systems do not account for variations in electrical loading and ambient environmental conditions. Thus, alarms do not always correspond directly to genuine operational issues. Online monitoring alone, and reliance solely on measurement data, is not sufficient. Monitoring data may show measurable changes in the density of SF6 within the chamber, but these changes are not always indicative of faults. In other cases, however, they may reflect more serious problems, such as gas leakage, which has both environmental and financial consequences. This study proposes a hybrid model that combines a physics-based prediction with a regression-based residual model. The proposed model can accurately predict monitoring output by combining sensor behaviour with a physical understanding of the system. This allows for changes in density to be interpreted more accurately, resulting in more meaningful anomalies being distinguished from normal operational variation. This, in turn, would enable faster detection of true anomalies while reducing needless interventions.

ORCID iDs

Brash, Alistair ORCID logoORCID: https://orcid.org/0000-0003-1370-1132, Shi, Guo ORCID logoORCID: https://orcid.org/0000-0002-6945-3150, Pestana, Jennifer ORCID logoORCID: https://orcid.org/0000-0003-1527-3178, Brown, Blair ORCID logoORCID: https://orcid.org/0000-0002-4734-9985, Tachtatzis, Christos ORCID logoORCID: https://orcid.org/0000-0001-9150-6805, Liu, Ting ORCID logoORCID: https://orcid.org/0000-0003-0604-4939, Irwin, Fiona, Lumsden, Nicol and Stephen, Bruce ORCID logoORCID: https://orcid.org/0000-0001-7502-8129;
CORE (COnnecting REpositories)