Learnings
Outcomes
The CASA project has delivered all core technical objectives and generated a comprehensive dataset and analysis suite that significantly advances the industry's understanding of PD behaviour in SF₆‑alternative gases.
The project:
Fully met its aims and objectives;
Developed novel diagnostic tools, including simultaneous RF/IEC measurement and ML‑based classification;
Generated actionable insights relevant to network operators, manufacturers, and standards bodies;
Produced replicable test methodologies and calibration frameworks;
Provided a clear pathway for field deployment and standardisation.
The outcomes provide a strong foundation for operational adoption of alternative gases in GIS and directly support SSEN Transmission’s environmental, reliability, and regulatory commitments. The project successfully increased the TRL level from TRL 2 to TRL 4. See section 10 for opportunities to develop the learning further.
Detailed deliverable progress reports from Cardiff University are available upon request. Significant new learning from the project has been delivered and published in technical papers and presented at international industry conferences:
Paper 1
A. Reid, R. Ullah, A. Haddad, M. Barnett, M. Nambiar and P. Taddei. “Partial Discharge Measurement in SF6-Alternative Electrical Insulation System.” CIGRE, 2024.
Paper 2
R. Ullah, A. Reid, M. Michelarakis, K. Zhang, A. Haddad, M. Barnett, M. Nambiar and P. Taddei. “Enhanced Partial Discharge Evaluation through Integrated RF and IEC Measurements.” ICD 2024, Toulouse France.
Paper 3
R. Ullah, A. Reid, S. Singh, A. Haddad, M. Barnett, M. Nambiar and P. Taddei. “Measurement of Surface Discharge Behaviour of Epoxy Insulation in Technical Air/ CO2 /C4-FN Mixture under Highly Divergent Electric Field Conditions.” ISH 2025 Karuizawa Japan.
Paper 4
R. Ullah, A. Reid, A. Haddad, M. Barnett, M. Nambiar and P. Taddei. "Feature Extraction and Classification of Partial Discharge Signals in C 4 F 7 N-Based Gas Insulated Systems: A Time-Domain and Machine Learning Approach." IEEE Transactions on Dielectrics and Electrical Insulation (2025).
Paper 5
R. Ullah, A. Reid, A. Haddad, M. Barnett, M. Nambiar and P. Taddei. "Enhancing Partial Discharge Diagnostics in C4-FN-Mixture, CO2 and Technical Air: A Simultaneous Measurement Approach." Journal Paper
Paper 6
R. Ullah, A. Reid, A. Haddad, M. Barnett, M. Nambiar and P. Taddei. “UHF-Based Partial Discharge Diagnostics in SF6-Alternative Gases.“ CIGRE 2026
Lessons Learnt
The project has generated several important lessons that will enhance the design and delivery of future projects, particularly those concerned with the transition to SF₆‑alternative insulation systems and the development of advanced diagnostic techniques for Partial Discharge (PD) monitoring.
1. Early Systems Integration Is Essential for Multi‑Sensor Measurement Approaches
The project demonstrated that simultaneous RF, IEC 60270 and UHF measurements provide highly complementary information. However, the integration of these systems late in delivery resulted in additional engineering effort and re‑calibration cycles.
Future Learning: Multi‑sensor measurement architectures should be defined, designed, and validated at the outset, with clear data synchronisation, triggering, and phase‑reference strategies embedded in early-stage planning.
2. Calibration Activities Require Greater Time, Resourcing and Iteration
Calibration of HFCTs, UHF sensors, and GTEM cells proved more complex than originally anticipated, particularly due to chamber resonances and sensitivity to geometric variations.
Future Learning: Projects utilising high‑bandwidth measurement systems must allocate dedicated time, specialist expertise, and formal test procedures for calibration and verification. Multiple calibration cycles should be assumed during planning.
3. Realistic Defect Models Are Critical to Producing Representative Diagnostic Data
Small variations in defect geometry (e.g., particle size, protrusion radius, electrode spacing) produced significant differences in PD behaviour. Simplified or idealised defects risk generating diagnostic signatures that diverge from real‑world conditions.
Future Learning: Where possible, defects should be manufactured to replicate realistic GIS/GIL failure modes, informed by asset data, OEM insights and field experience.
4. Alternative Gas Mixtures Cannot Be Interpreted Using SF₆‑Based Diagnostic Assumptions
The project confirmed that C₄F₇N/CO₂ mixtures, technical air, and other alternative gases exhibit distinct PD characteristics, behaviours and spectral signatures.
Future Learning: Future diagnostic schemes for alternative gases must be developed independently of SF₆ knowledge. Gas‑specific diagnostic thresholds, classification rules and monitoring strategies are required.
5. PRPD Analysis Alone Is Insufficient for Reliable Classification Across All Defects
While PRPD patterns provided clear insight for several defect types, certain topologies (e.g. free metallic particles) produced overlapping or ambiguous patterns.
Future Learning: Multi‑domain analysis—combining PRPD, time‑domain information, spectral analysis and statistical features—should be used where defect discrimination is safety‑critical.
6. UHF‑Only Monitoring Is Viable but Requires Advanced Feature Engineering
The project demonstrated that UHF sensors can classify defects without phase-reference when combined with frequency‑domain energy extraction and statistical classification.
Future Learning: Online monitoring systems should embed feature‑engineering algorithms rather than rely on simple pulse counting or amplitude‑only metrics.
7. Machine Learning (ML) Techniques Provide Value but Depend on High‑Quality, Labelled Data
ML models delivered strong performance; however, they required extensive data cleansing, feature selection and robust labelling.
Future Learning: When ML is included within project scope, data acquisition strategies should be designed with ML model development in mind, including noise management, data quality assurance and traceable labelling processes.
8. Test Chamber Geometry Strongly Influences UHF Results and Must Be Controlled
Energy distributions were sensitive to sensor placement and chamber resonances.
Future Learning: Sensor locations, chamber geometries and boundary conditions must be documented rigorously and remain consistent throughout testing to ensure reproducibility and comparability across test phases.
9. Field Validation Should Be Positioned Earlier in Project Delivery
While the project produced strong laboratory evidence, in‑service validation was identified as future work.
Future Learning: Future projects could include structured field trials within the main delivery plan to validate laboratory findings under operational conditions, including long‑term monitoring.
10. Engagement With Standards Bodies Should Be Initiated Early
The project produced outcomes relevant to IEC 60270 and IEC 62478 but engagement with standards bodies occurred at the end of the project.
Future Learning: Early coordination with IEC, CIGRE and OEM partners accelerates the translation of research findings into usable guidance and ensures applicability to existing industrial practices.
11. Quantification of Environmental and Network Benefits Strengthens Strategic Value
The project provided qualitative evidence of environmental benefits via SF₆ reduction.
Future Learning: Future work should incorporate quantified assessments of greenhouse gas savings, defect detection improvements and potential reductions in outage risk to support investment cases and regulatory justification.
12. Structured Stakeholder Engagement Enhances Technical Quality and Industry Acceptance
Collaboration with academic and industrial partners proved valuable.
Future Learning: Establishing formal stakeholder groups, including academics, OEMs and other TOs, supports knowledge sharing, challenge scrutiny, and alignment with wider industry practice.
