Partial discharge (PD) is localized dielectric breakdown that does not bridge the full insulation gap — a single event is harmless, but thousands per day progressively erode paper insulation through carbonized tracking ("treeing"), ultimately causing turn-to-turn or layer-to-layer short circuits. Three complementary detection methods exist: electrical (pC measurement, quantitative, highest sensitivity, typically offline), acoustic (20-100 kHz ultrasonic, ~10 cm triangulation accuracy, online-capable), and UHF (300 MHz-3 GHz, excellent noise immunity, online). The gold standard for localization is the acoustic-electrical combined method, which uses the electrical signal for microsecond-precision triggering and acoustic time-difference-of-arrival for 3D spatial location. DGA — specifically rising H₂ trends — provides a continuous PD proxy, though online monitor reliability remains a field concern.
Engineering note: values and thresholds in this article are reference points for screening and discussion. Final acceptance should follow the project specification, applicable IEC/IEEE standards, local utility requirements and the approved factory test protocol.
Partial discharge is the localized dielectric breakdown of a small portion of the electrical insulation — usually occurring in a gas-filled void, at an oil-paper interface, or along a contaminated surface. Critically, PD does not bridge the full insulation gap between conductors. The transformer continues to operate, passing all routine electrical tests. This is what makes PD so insidious: the insulation is failing, but in slow motion.
A single PD event transfers picocoulombs of charge and releases microscopic energy — on its own, entirely harmless. But in a transformer with active PD, thousands of discharge events occur every day, concentrated at the same defect site. Each discharge:
Over months to years, the carbonized trees grow until they bridge the full insulation thickness. At that point, a single discharge event becomes a full short circuit — and the transformer fails, often without warning.
PD does not distribute uniformly through the winding. It concentrates in the highest-stress regions:
No single PD detection method is universally best. Each method has a distinct sensitivity profile, operational requirement, and cost point. Selecting the right method — or combination of methods — depends on whether the transformer is at FAT (factory acceptance testing), in-service, or under investigation after a DGA alarm.
| Parameter | Electrical (IEC 60270) | Acoustic | UHF |
|---|---|---|---|
| Measured quantity | Apparent charge (pC) via coupling capacitor or bushing tap | Ultrasonic pressure wave (20-100 kHz) via piezoelectric sensors on tank wall or in oil | Electromagnetic wave (300 MHz-3 GHz) via internal or external UHF antenna |
| Sensitivity | Highest — 1-10 pC detection threshold | Moderate — depends on acoustic path; attenuated by barriers | High — 5-50 pC equivalent in well-designed installations |
| Online vs. offline | Primarily offline (energized by external test source); on-line possible via bushing taps but noise-limited | Online — sensors attach magnetically to tank wall externally; no outage required | Online — antennas installed through oil drain valves or dielectric windows; no outage required |
| Localization accuracy | Poor — can confirm PD is present but cannot locate its source within the winding | Good — triangulation with 3-4 sensors gives ~10 cm accuracy. Sound velocity: ~1400 m/s in oil, ~4000 m/s in solid insulation | Moderate — TDOA measurement across multiple antennas; accuracy ~20-50 cm |
| Noise immunity | Poor — susceptible to corona from bushing surfaces, nearby switchyard equipment, and power-line carrier signals | Moderate — immune to electrical noise but susceptible to mechanical noise (cooling fans, rain, core magnetostriction) | Excellent — most external corona and switching noise is concentrated below 300 MHz; UHF band is naturally quiet |
| Quantitative (pC calibrated) | Yes — the only method that directly measures apparent charge per IEC 60270 | No — amplitude correlates with PD magnitude but cannot be calibrated to pC | No — signal amplitude depends on antenna location and PD-source geometry |
| At FAT | Standard — required by IEC 60076-3 for transformers ≥ 72.5 kV | Supplementary — used if electrical PD exceeds acceptance criteria and source must be located | Supplementary — increasingly common for large power transformers during FAT |
| In-service (online) | Limited — bushing tap measurement possible but noise floor is high | Practical — periodic survey or permanent installation; acoustic sensors are economical | Practical — permanent UHF antenna installation for critical transformers |
| Approximate relative cost | $$ (test set + trained operator) | $ (portable instrument, periodic survey) | $$$ (permanent antenna installation, monitoring system) |
Per IEC 60270, the electrical method uses a coupling capacitor and measurement impedance to detect the high-frequency current pulses generated by each PD event. The measured quantity is apparent charge in picocoulombs (pC). This is the only method that produces a calibrated, quantitative, internationally standardized measurement.
The acceptance criterion for new transformers, per IEC 60076-3, is <300 pC at 1.5 times rated voltage, with no rising trend during the test. Utilities and industrial users with higher reliability requirements often specify stricter limits — 100 pC or even 50 pC — particularly for generator step-up (GSU) transformers and critical substation transformers.
The limitation of electrical measurement during factory testing is that it tells you PD is present but not where. If the acceptance criterion is exceeded, a localization method — typically acoustic — must be employed to determine whether the source is in the winding (design or manufacturing defect), at a bushing (component defect), or at a bushing current transformer (assembly issue).
The acoustic-electrical combined method is the most accurate technique for pinpointing a PD source within a transformer. It leverages the best properties of each individual method: the microsecond timing precision of the electrical signal and the spatial information carried by acoustic waves.
Under favorable conditions (good acoustic paths, minimal internal obstructions), the combined method achieves localization accuracy of approximately 10 cm. This is sufficient to distinguish between PD in the winding (design defect), at a bushing interface (assembly defect), or at a bushing current transformer (component defect) — three scenarios with very different corrective actions.
When partial discharge occurs in an oil-filled transformer, the discharge energy breaks carbon-hydrogen bonds in the mineral oil molecules. The simplest and most abundant decomposition product is hydrogen (H₂). Unlike thermal faults — which produce a spectrum of hydrocarbon gases (CH₄, C₂H₄, C₂H₆) — PD is characterized by a dominant H₂ rise with minimal or no corresponding hydrocarbon increase.
This makes rising H₂ the most sensitive DGA indicator of developing PD activity. The DGA three-ratio method (C₂H₂/C₂H₄ → CH₄/H₂ → C₂H₄/C₂H₆) distinguishes PD from thermal faults: a CH₄/H₂ ratio below 0.1 is characteristic of PD, while higher ratios indicate thermal degradation.
Online dissolved gas analysis monitors have transformed transformer condition monitoring by providing continuous gas-in-oil data without the cost and delay of periodic laboratory sampling. However, field experience reveals significant reliability limitations.
Data from a large-scale deployment — 2,164 online DGA monitors installed across one Chinese provincial grid — showed that approximately one-third of the monitors produced unreliable readings. Failure modes include:
Use online DGA for trending, not precision measurement. A steadily rising H₂ concentration — even from an imperfectly calibrated monitor — provides early warning of developing PD that warrants investigation. Confirm with laboratory DGA quarterly. A divergence between online and lab results is itself diagnostic: it signals that the online monitor requires maintenance (sensor replacement, membrane cleaning, or recalibration).
Laboratory DGA using gas chromatography remains the gold standard for accuracy and the definitive basis for any condition assessment or maintenance decision.