A paper at the 2026 CIGRE Session (A2_11489_2026, HD Hyundai Electric) places PD sensors using a modelled attenuation matrix and localises a source with three sensors instead of four, reporting a 20 % accuracy improvement on an operating transformer. A 2026 finite-difference time-domain study of a 110 kV transformer adds numbers: the best single sensor position covered 67 % of the tank, a diagonal two-sensor arrangement 97 % (J. Phys.: Conf. Ser. 3229). Both argue about sensor placement. The choice between online vs offline partial discharge testing is a different question, decided by two standards that never quite meet.
Two standards, two measurement worlds
Offline testing is conventional measurement in the sense of IEC 60270. The unit is de-energised, a calibrated pulse is injected, and the answer comes back in picocoulombs — a traceable, absolute quantity comparable to the routine test on the nameplate.
Online testing is non-conventional measurement in the sense of IEC TS 62478, adopted in China as GB/T 42287. The unit stays in service: a UHF coupler at an oil valve, an HFCT around a ground connection, an acoustic sensor on the tank wall or an RF probe picks up what the source emits, and the reading is relative — a voltage, a dB figure, a count rate — read as a trend against that sensor's own history. DL/T 1498.1 governs the on-line device. One boundary is easy to get wrong: IEEE C57.127 covers acoustic emission only and cannot be cited for a UHF channel.
Offline versus online, side by side
| Offline (conventional, IEC 60270) | Online (non-conventional, IEC TS 62478 / GB/T 42287) | |
|---|---|---|
| Equipment state | De-energised, disconnected, grounded as required | In service, at whatever load the system carries |
| Quantity reported | Apparent charge in pC, pulse-injection calibrated | Amplitude in mV or dB, plus rate and phase data |
| Calibration | Traceable pulse injection; comparable to factory tests | No pC traceability; each sensor read against its own baseline |
| Duration | A snapshot — minutes to hours | Continuous, over months and years |
| Best at | Commissioning baselines, post-repair checks, gross defects | Faults that appear only under real load, temperature and time |
| Blind to | Anything not active during the test; no thermal or load stress | What the coupling path attenuates; noise; shadowed sources |
FAQ: Is online partial discharge testing as accurate as offline?
Not in the same units, and the online vs offline partial discharge testing comparison is not a fair one. Offline measurement is accurate in picocoulombs because it controls the coupling path and calibrates it before the test. Online measurement trades that absolute scale for continuous presence: the path runs through a tank wall, an oil valve or a ground strap, and cannot be pulse-calibrated while the unit carries load. What you gain is a fault that exists only at operating temperature, seen on the day it starts rather than the day of the next outage.
FAQ: Can an online PD monitoring system be calibrated in picocoulombs?
Not to the traceability of IEC 60270, and a pC figure offered from an in-service UHF or acoustic channel is an estimate, not a calibration. Establish a baseline while the unit is known good, then treat change as the signal: a channel flat for two years that has climbed for three weeks is reporting something, whatever the absolute number on the display. A quoted pC figure for an online channel is vendor data tied to a stated setup, not a measurement uncertainty budget.
When each method is the right one
They are complements. Offline testing belongs at commissioning, after a repair or a bushing change, and whenever you need a number comparable to a factory test — it is the only route to a pC-traceable baseline. Online monitoring belongs on units that cannot be taken out on demand: generator step-up transformers, large grid transformers, and any unit whose outage costs more than the monitoring itself. Overlaid, the offline snapshot dates the baseline and the online channel watches it drift.
That is why the online vs offline partial discharge testing decision rarely comes down to sensitivity, but to which units you can afford to switch out and how much of the tank one sensor actually covers.
Designed in line with IEC TS 62478:2016 and GB/T 42287, the TPD-400 online PD monitoring system takes UHF (300–1500 MHz at an oil valve or manhole), RF, acoustic (80–200 kHz) and HFCT (0.3–30 MHz) inputs into one industrial IED, with 125 MS/s 14-bit acquisition, channel synchronisation under 100 ns, minimum measurable apparent charge around 5 pC and 600 kV lightning-impulse withstand (vendor data). Pattern classification runs as a recognition expert system, and UHF time-of-flight is offered as an aid to locating a source, not a guarantee. Reporting runs over IEC 61850 (DL/T 860). The same unit reads a tank wall alongside an oil-valve UHF retrofit and pairs with a DGA-900 nine-gas analyser — see PD versus DGA and utility-wide monitoring.
Talk to us about which units belong on an online channel.
Sources
- CIGRE Session 2026, A2_11489_2026 (HD Hyundai Electric; MIN Byoung-Woon, LEE Danbi, BAE Kwang-Don, LEE Jeong-Bok): digital-twin PD diagnostics, attenuation-matrix sensor placement, localisation with three sensors versus four, 20 % accuracy improvement on an operating transformer.
- Journal of Physics: Conference Series 3229 (2026): FDTD study, 110 kV three-phase transformer — single-sensor coverage 67 %, diagonal two-sensor 97 % at the same threshold.
- CN 122410222 A, 17 July 2026 (China Electric Power Research Institute): converter-transformer copper shielding plates used as RF sensing antennas.
- IEC 60270 (conventional, offline PD measurement); IEC TS 62478:2016 (non-conventional, online; adopted as GB/T 42287); DL/T 1498.1 (on-line monitoring device requirements); IEEE C57.127 (acoustic emission only); IEC 61850 (DL/T 860). IEEE PES T&D 2026, Chicago, 4–7 May 2026.
- TPD-400 sensing and acquisition figures: vendor data.