October 30, 2026 · Fault Diagnosis

In January 2026 a one-day IEEE DEIS workshop in Malaysia was given over entirely to the gap between partial discharge theory and practice under IEC 60270. Its agenda — plot reading, offline versus online measurement, live demonstration — names a problem the standard does not address: two engineers can follow IEC 60270 completely and still report different numbers for the same defect. Interpreting PD patterns therefore starts before the plot is drawn, with the settings that produced it.

What the phase-resolved plot actually is

A phase-resolved partial discharge (PRPD) plot folds every detected pulse into one cycle of the supply waveform. The horizontal axis is phase, 0 to 360 degrees; the vertical axis apparent charge; colour or height carries pulse count. Time disappears, and that is the point — a defect repeating at the same place in every cycle forms a shape, while random interference smears across all phases.

The shape is only meaningful against a real phase reference. On a conventional IEC 60270 circuit the reference comes from the test arrangement; on non-conventional methods under IEC TS 62478 — UHF, HFCT, acoustic — it has to be reconstructed. The tolerance is tighter than it sounds: at 50 Hz one millisecond of synchronisation error is already 18 degrees of phase shift.

Instrument settings move the number by more than 20 per cent

IEC 60270 fixes the measurement principle, not the signal chain. Bandwidth, filtering, pulse resolution, integration window and evaluation logic are left to the user inside permitted ranges. A 2026 technical report from HIGHVOLT and Red Pitaya quantified the cost of that freedom: varying these settings within what the standard allows produced more than 20 per cent variation in measured QIEC from a single artificial source (vendor data).

That is wider than most acceptance margins. Limits such as under 10 pC for HV extruded cable and under 20 pC for HV test transformers, quoted in published specifications, can be decided by a filter setting nobody recorded. Store bandwidth, filter, resolution and calibrator date with every capture, and never compare plots whose settings differ.

Where shape alone stops working

A 2026 thesis from Middle East Technical University tested corona, surface, internal and interfacial discharge and reached a blunt conclusion: PRPD alone does not always separate discharge mechanisms, and phase-only reading invites misjudgement. The same work found surface discharge to be continuous with corona rather than a separate category — a distinction most pattern atlases present as clean.

Two further limits matter. In natural esters, pulse magnitudes and repetition rates differ from mineral oil, so apparent-charge thresholds borrowed from mineral-oil practice do not transfer; a 2026 study on vegetable oils needed DBSCAN clustering to separate metallic-particle faults from dispersed moisture. On HVDC equipment there is no periodic reference to fold against at all, so pulse-sequence analysis replaces PRPD rather than supplementing it.

Added quantity What it contributes When it decides the call
H(Q) distribution Pulse-amplitude statistics and skew Two defects share a phase region
Leakage current Independent surface condition signal Pressboard or bushing involvement
Three-phase comparison Separates internal defects from common-mode noise Any suspect phase offset
UHF or acoustic channel Different mechanism, different noise floor Inverter or thyristor interference dominates
Dissolved gas trend Chemical consequence over hours to days Deciding whether to act now

One published case makes the last row concrete: where thyristor interference pushed the charge-based channel past 3 nC and toward 7.8 nC (published case data), IEC 60270 PRPD was unusable while UHF stayed readable. The method did not fail; the environment did.

Where the plot and the gas disagree

Interpreting PD patterns gives evidence about a mechanism; dissolved gas analysis gives evidence about its consequences — the division of labour is set out in PD versus DGA monitoring. A plot showing activity against a flat gas trend suggests a defect not yet producing measurable decomposition. Gas rising without a matching plot suggests a source the coupler cannot see.

How often should PD patterns be re-read?

At the same interval as your alarm review, and always after any change to instrument settings, coupler or phase reference. A pattern whose baseline was captured under different filters is not a trend — it is two unrelated measurements. Treat a settings change as a new baseline and record it as one.

Can software identify the defect for me?

Software can speed up interpreting PD patterns, but clustering and machine learning sort noise from signal — they consume a clean measurement and do not replace the phase reference or the calibration chain behind it. Use them as a second opinion, not a substitute.

An online monitor is judged on the same three checks. The TPD-400 synchronises channels to under 100 ns (vendor data), keeping inter-phase offsets small enough to read. It is designed in line with IEC TS 62478 and GB/T 42287 and communicates over IEC 61850 (DL/T 860); for the charge-based reference measurement, see what IEC 60270 compliance requires. Pair it with a gas channel and the two limits cover each other: DGM-900 records seven gases and moisture on the same transformer, so a plot without a gas trend becomes visible as a disagreement rather than as silence. Pattern families are covered in PRPD and PRPS analysis, alarm logic in PD alarm thresholds. Send the settings and phase reference with your data through our contact form and we will tell you what the plot can support.