Unplanned outages caused by internal arc faults remain one of the most critical risks in Gas Insulated Switchgear.
While conventional protection systems can clear faults rapidly, utilities are often left with limited information on where the fault occurred in GIS, extending outage duration and restoration time. Recent advances in UHF PD condition monitoring now enable automatic arc detection and localisation, delivering actionable intelligence within seconds of an event.
The Challenge of Arc Faults in GIS
Arc faults in GIS occur when the dielectric strength of SF₆ or alternative insulating gases is compromised due to defects, ageing, contamination, or transient overvoltage. Unlike operational arcs generated during normal switching, fault arcs cause severe thermal, mechanical, and electrical stress, requiring immediate interruption. However, identifying the exact faulted section after protection operation remains a challenge, especially in complex substations with multiple bays and busbars.
Faster and more precise fault localisation directly translates into reduced outage time, targeted inspections, and safer re-energization of healthy assets.
Combining Arc Detection with Continuous PD Monitoring
UHF Partial Discharge monitoring is a proven technique, widely adopted for condition-based maintenance of GIS. The same UHF sensors used for PD detection are also capable of detecting arc emissions. Recent developments demonstrate for the first time, a fully integrated system that continuously detects, classifies, PD activity and arc faults detection and localization using the same hardware infrastructure.
This approach eliminates the need for dedicated arc sensors or optical systems and enables retrofitting on energised GIS with minimal effort.
Intelligent Arc Classification Using Machine Learning
A key advancement lies in the application of machine learning to distinguish arc signals from other high-energy events. Two levels of supervised classifiers are used:
Level 1: Differentiates arc signals from non-arc signals such as PD, noise and communication interference.
Level 2: Further classifies arc signals into fault arcs and normal switching arcs from circuit breakers.
These models were trained using millions of labelled signal patterns, combining laboratory-generated arc data with real-world signals collected from an extensive global install base. Cross-validation results demonstrated classification accuracies above 98%, ensuring high confidence in operational environments.
Automatic Arc Localization by Relative Amplitude
Once an arc is detected and classified, the system automatically locates the source using relative signal amplitude across multiple UHF sensors. Because UHF signal attenuation increases with distance and varies with GIS components, the sensor detecting the highest amplitude provides a strong indication of proximity to the source.
To enable this, the GIS attenuation profile is mapped during commissioning using standard injection techniques. The system then correlates real-time signal amplitudes with this model to estimate the most likely arc location.
Unlike traditional time-of-flight methods which require expert engineers and site visit. This process runs continuously, automatically, and in real time, using installed monitoring hardware.
Proven Performance on Live GIS
On-site testing on a live 420 kV GIS demonstrated the effectiveness of this approach. Simulated arc signals were injected at various locations, including bays, busbars, and coupler sections. In every test case, the system successfully localized the fault to the correct half-bay, meeting the accuracy required for practical restoration planning.
This level of precision enables operators to immediately narrow the faulted zone such as feeder side versus busbar side of a breaker, significantly reducing investigation time.
Delivering Actionable Intelligence
By combining continuous UHF monitoring, machine learning-based classification, and automatic localisation, utilities gain a powerful tool for managing internal GIS faults. The result is faster fault isolation, improved operator confidence, and reduced outage impact, turning arc events from disruptive failures into manageable incidents.
This article was originally published in the May 2026 issue of the Reliability Engineered Design magazine.
View Magazine
