Partial discharge (PD) is a phenomenon that occurs in electrical assets whereby localised energy discharges do not bridge the insulation gap between conductors. It is a sign that an asset has a defect of some kind, either from manufacture or deterioration. Initially PD may occur infrequently, occurring more often as the defect worsens until eventually the PD frequency and magnitude becomes sufficiently concerning that the asset needs to be replaced or the discharge bridges the gap causing arcing and failure. PD monitoring is carried out routinely every three months along with thermovision checks at each substation; it is time consuming, and the monitoring equipment is relatively heavy. More frequent PD checks would increase the probability of getting early warnings of asset deterioration but other than more frequent personnel checks the alternative is continuous monitoring systems that have limitations and are expensive. In the event of a dielectric failure of an asset it is sometimes necessary to establish risk management hazard zones, PD monitoring could be used to mitigate the risks and allow site work to continue.
This project will address the problem in a number of different ways to improve our ability to identify, locate and diagnose PD related defects as early as possible. The project will involve a number of workstreams:
- Use of machine learning techniques to identify the optimal locations for PD monitoring systems
- Improved diagnostic capability for understanding PD patterns taking into account influencing environmental factors using AI
- Demonstration of robot- and drone-mounted PD monitoring – assessing the capabilities and relative merits of both solutions
Benefits
The project will deliver improved monitoring and can prevent failure, and identify causes earlier thereby reducing outages or allowing them to be aligned with construction activity
Robots and drones could provide mitigation after a failure as they can enter risk management hazard zones. This could enable work to continue short term, continuous monitoring can allow long term-working.
Learnings
Outcomes
WS2
The study of monopole antenna configurations to enhance PD detection through improved impedance matching and signal reception showed that a combination of two antennas had superior SWR performance and consistently captured higher PD signal energy when validated using a handheld PD device. Connector type was also found to influence results, with elbow-type outperforming goalpost connectors. These findings support the integration of optimized multi-antenna setups in handheld PD instruments to improve measurement accuracy and diagnostic reliability in high-voltage systems.
Recommendations for further work
None at present.
Lessons Learnt
WS1
Localisation of PD using ‘fingerprinting’ compares measured signal strength to a pre-recorded dataset to determine the location of a signal, often using the k-nearest neighbours (KNN) algorithm. Artificial neural networks (ANN) are an alternative solution but require extensive training data. Early results suggest that the ANN approach offers an improvement in localisation accuracy.
WS2
Experimental results in the laboratory have provided insights into how antenna configurations influence impedance matching and signal reception in broadband radio frequency (RF) applications, particularly for handheld PD detection. Individual analysis showed that while the commercially available wideband monopole antennas bought for this project offer reasonable standalone performance, they can suffer from poor impedance matching across much of the spectrum and resulted in higher standing wave ratios (SWR). One of the antennas demonstrated better matching at lower frequencies but its overall bandwidth coverage was limited.
The most compelling results emerged from dual-antenna configurations, which demonstrated significantly reduced SWR values and enhanced impedance stability across a broader frequency range. This translated into improved PD signal capture performance, as validated by energy analysis from laboratory PD measurements.
WS4
When the robot moves under energised busbars, a small charge is induced in the legs as they are lifted from the ground. When placing them down again, the charge is dissipated. The PD locator is sensitive enough to detect this discharge.
Dissemination
A representative from SPEN has joined project review meetings for WS1, WS2 and WS5. Representatives from SSEN-T have joined meetings for WS1.
This project will be a case study in the National Grid Annual Innovation Summary for 2024/25
WS1
A presentation “Neural Network for Partial Discharge Localisation in Simulation” is planned for the Universities High Voltage Network conference in Liverpool in June 2025.
A paper entitled “A 3D Ray Tracing Based Model to Improve Substation-Wide Partial Discharge Localisation Accuracy” has been submitted to the International Symposium on High Voltage Engineering (ISH) for presentation in Nagano, Japan in August 2025.
WS2
A paper entitled “Optimizing Small Antenna Configurations for Partial Discharge Monitoring of Electrical Assets” has been submitted for to the IEEE Conference on Electrical Insulation and Dielectric Phenomena to be held in Manchester in September 2025.
WS5
A presentation on WS5 was given at the Eurodoble conference in Manchester in September 2024.
