Partial Discharge Measurement — Essentials, Pro and Masterclass

Apparent charge is the external echo, not the local damage.

Partial discharge is a localised electrical discharge that only partially bridges the insulation between conductors. It occurs where the local electric field exceeds the local dielectric strength: inside a gas cavity trapped in resin, at the interface between a badly fitted stress cone and cable insulation, at the tip of an electrode in air, under a damaged semiconductive coating on a stator bar. Each individual event involves a few picocoulombs to a few nanocoulombs of apparent charge and lasts nanoseconds to a few hundred nanoseconds — a minuscule energy, repeated tens or hundreds of times per voltage cycle, millions of times per hour of operation.

The diagnostic relevance comes from the asymmetry between cause and effect. PD is almost never the defect. It is the early symptom of a manufacturing, assembly or ageing defect, and simultaneously an accelerating agent of degradation, because every discharge bombards the cavity surface with electrons, ions and chemical by-products. This module turns the routine pass/fail criterion into an investigation: why the discharge is born, what apparent charge actually represents, how the phase-resolved pattern names the defect, and how a complete measurement campaign is run from acquisition to maintenance decision. The structure follows three layers — Essentials covering physics and circuit, Pro covering limits, patterns, noise and non-conventional methods, and the Masterclass working three real cases from noise to verdict.

What you will be able to do

• Explain PD physics — internal cavities, corona, surface discharge and floating electrode — and the specific degradation mechanism in each insulation technology: oil-paper, XLPE, epoxy, mica, SF6.

• Build and calibrate the measurement circuit per IEC 60270: coupling capacitor, measuring impedance, traceable calibrator, bandwidth selection and performance verification.

• Distinguish apparent charge from real charge and interpret q, n, inception voltage and extinction voltage in the context of each product standard's voltage procedure.

• Recognise typical PRPD patterns and separate PD from noise by filtering, gating and source separation tools such as three-frequency clustering and time-frequency mapping.

• Conduct a complete campaign — acquisition, cleaning, clustering, diagnosis and maintenance decision — in laboratory, factory and field, using the conventional charge-based method and the electromagnetic and acoustic methods of IEC TS 62478.

Essentials I — where the discharge is born

The classic classification distinguishes four families of PD, each with its own physics, signature and severity. Internal discharge occurs in gas cavities trapped inside solid insulation or in delaminated interfaces. Because the permittivity of gas is about 1 while the solid is 2.3 in XLPE and 3.5 to 4.5 in epoxy and impregnated paper, the field inside the cavity is amplified by the permittivity of the surrounding material, while the dielectric strength of the gas — about 3 kV/mm in air at atmospheric pressure against tens of kV/mm in the solid — is far lower. The cavity breaks down long before the sound insulation does.

The most widespread teaching model represents the object with three capacitances: the sound insulation in parallel, the insulation in series with the cavity, and the cavity itself. When the voltage across the cavity reaches the ignition value of the gas, the cavity discharges in nanoseconds, the cavity voltage collapses, and the voltage step is redistributed through the capacitive divider, producing a current pulse at the object terminals whose integral is the apparent charge. Because the series capacitance is far smaller than the parallel one, the charge circulating at the terminals is much smaller than the charge actually transferred in the cavity — typically one to two orders of magnitude smaller. Apparent charge therefore does not measure local damage. It measures the external echo of the event, and it is only comparable between objects of the same geometry and capacitance.

Essentials II — the measurement circuit and its calibration

The world reference for conventional measurement is IEC 60270. The edition used in most laboratories and cited by product standards is IEC 60270:2000+AMD1:2015. In June 2025 the fourth edition was published under a revised title, charge-based measurement of partial discharges, formally extending the scope to alternating voltages up to 500 Hz and to direct voltage, and consolidating charge-based measurement as a category distinct from the electromagnetic and acoustic methods. New procedures should reference the fourth edition, while reports issued against product standards citing earlier editions remain valid in that context.

The PD pulse at the terminals has a spectrum extending to tens of megahertz, but the integral — the charge — is contained in the low-frequency components. IEC 60270 defines wideband systems with lower cut-off between 30 kHz and 100 kHz, upper cut-off up to 1 MHz, and bandwidth between 100 kHz and 900 kHz; and narrowband systems with bandwidth of 9 kHz to 30 kHz and centre frequency between 50 kHz and 1 MHz. Wideband preserves polarity and resolves close pulses; narrowband makes it easier to dodge AM broadcast stations in the field, at the cost of losing polarity and broadening the pulse. Two performance parameters are critical: pulse resolution time, on the order of 10 µs or less, since pulses closer together than that overlap and produce reading errors, and pulse train response, specified to guarantee correct reading of the standardised quantity, which is the largest repeatedly observed value rather than an isolated peak.

Pro I — limits by family, and why they are not comparable

Product standards prescribe voltage profiles shaped like a house: rise to an enhancement voltage, hold briefly to condition any cavities, then drop to the measurement voltage where PD is recorded for a defined time. For a liquid-immersed power transformer under IEC 60076-3, the full profile is background at or below 0.4·Ur/√3, then 1.2·Ur/√3, a level at 1.58·Ur/√3, enhancement at 1.8·Ur/√3 for 60 s, a measurement level of 60 minutes at 1.58·Ur/√3, and a final check at 1.2·Ur/√3 — with criteria of 250 pC or less during the hour, no rising trend with an increase of 50 pC or less over the hour, and 100 pC or less at the final level. Background must be 50 pC or lower for the test to be valid. Watch a classic trap here: the pair 1.8 and 1.3 belongs to dry-type transformers under IEC 60076-11, with a typical 10 pC limit, and must not be transplanted to liquid-immersed units.

Two observations structure a professional reading of that list. First, the picocoulomb values are not comparable across families. 250 pC in a transformer is not worse than 10 pC in a current transformer, because capacitance, geometry and insulation technology differ. The physical reason is in the material. Oil-paper generates dissolved gases and waxes that dissolved gas analysis tracks, and the oil partially heals the interface, so degradation is slow and monitorable — hundreds of pC are tolerated in test. XLPE and EPR erode locally into electrical trees with no self-repair, and the time from tree inception to breakdown can be days to months, so routine tests demand no detectable discharge above background. Cast epoxy erodes casting voids and delaminates at conductor interfaces with no self-repair, giving 5 to 50 pC limits. Mica-epoxy is exceptionally resistant, so bars live with nanocoulombs for decades and interpretation is by trend and pattern. SF6 decomposes into corrosive by-products and a free particle can migrate to a spacer and precipitate flashover, so any sustained activity is investigated.

Pro II — reading the pattern and separating the noise

The phase-resolved partial discharge map accumulates every pulse, cycle after cycle, in a phase-versus-amplitude plane. Within minutes, tens of thousands of events draw the statistical signature of the source. Four fundamental signatures anchor the reading. An internal cavity produces clusters on the rising flanks of both half cycles, approximately symmetric in phase and amplitude. Corona in air concentrates first at the peak of the half cycle in which the sharp electrode is the cathode, with regular pulses of nearly constant amplitude, and only at higher voltage does activity appear in the opposite half cycle. Surface discharge gives broad clusters with wide amplitude spread and clear asymmetry between half cycles. A floating electrode draws horizontal bands of nearly constant amplitude scattered in phase, often switching between discrete levels.

In factory and field, the enemy is noise: drive thyristors with pulses synchronous to the mains at fixed phase, AM broadcast carriers as continuous modulated sinusoids, external corona from the test arrangement itself, and intermittent contacts. The arsenal has layers. Linear filtering removes radio bands but broadens the pulse and degrades temporal separation, and it must be documented in the report because it changes the reading. Gating cuts time or phase windows where a known source emits, at the cost of blinding the system in that window — undocumented gating is evidence tampering. A reference antenna allows subtraction or vetoing of external events. And pulse-shape separation exploits the fact that each source travels its own electromagnetic path to the sensor and therefore arrives with its own spectral signature: time-frequency maps and synchronous three-frequency amplitude comparison group events into clusters, and each cluster redisplayed as an individual PRPD reveals a pure source. That is the difference between reporting 300 pC in the test and reporting a 300 pC cavity in phase B with the rest being busbar corona.

Masterclass — non-conventional methods and the field verdict

Charge-based measurement requires galvanic access and a controllable environment. When the equipment is in service, or when noise makes the IEC 60270 band unusable, the methods of IEC TS 62478 take over by capturing other physical manifestations of the same event. A high-frequency current transformer is a ferrite core with a through window installed on the shield or tank earthing conductor, with typical bandwidth from about 1 to 30 MHz. It is the workhorse of on-line PD in cables and transformers: it installs without an outage, on the earth conductor, with good sensitivity and the possibility of location by transit time.

Application in the field closes the loop. In factory routine testing a cable is short and the criterion is binary. In commissioning and maintenance the question changes from whether there is PD to how many metres away it is. The answer is reflectometry: a PD pulse born in a joint travels both ways at a typical velocity of 150 to 170 m/µs in XLPE, and the instrument at the termination receives the direct pulse and then its reflection from the far end, with the time difference giving the source position. At 160 m/µs and 20 ns of temporal resolution, theoretical spatial resolution is about 1.6 m, degraded in practice by dispersion and noise. Joints and terminations concentrate the overwhelming majority of field defects, because they are made by hand. To energise long cables in the field the options are resonant sources, VLF at 0.1 Hz with capacitive current 600 times lower than at 60 Hz, or damped AC. In every case, PD readings under VLF differ from those at power frequency — fewer cycles per second, different space charge dynamics, potentially different inception and extinction voltages — and the report must state waveform, frequency and duration.

Syllabus

• Taxonomy: internal cavity, surface discharge, corona, floating electrode

• The abc capacitive model, its modern critique, and Townsend versus streamer regimes

• How PD kills each insulation: oil-paper, XLPE and EPR, cast epoxy, mica-epoxy, SF6

• IEC 60270 editions: the 2000+A1:2015 edition and the 2025 charge-based edition

• Circuit components: test object, coupling capacitor, coupling device, blocking filter, integrating instrument

• Wideband and narrowband systems, pulse resolution time and pulse train response

• Calibration as metrological evidence, and when recalibration is mandatory

• Voltage procedures: the enhancement and measurement profile of each product standard

• Limits by equipment family, and why pC values are not comparable across families

• PRPD: the four fundamental signatures and the machine-specific and GIS-specific patterns

• Noise separation: filtering, gating, reference antenna, pulse-shape clustering

• The five-stage campaign: acquisition, cleaning, clustering, diagnosis, decision

• Non-conventional methods per IEC TS 62478: HFCT, UHF, acoustic

• Sensitivity verification for UHF systems in gas-insulated switchgear

• Application by equipment: cables with PD location, rotating machines, GIS and transformers

• The shielded laboratory: Faraday cage attenuation, achievable background, internal discipline

Laboratory work

The practical uses a test object with a known defect — a resin specimen with a calibrated cavity, or a deliberately badly assembled cable termination — and an injectable noise generator. The routine mirrors the ten-step setup checklist. Step 1: inspect the circuit for a PD-free source, a busbar without points, and short connections without loops. Step 2: verify background noise with the circuit energised at low voltage and record it in pC in the report. Step 3: select and connect a coupling capacitor suited to the object capacitance. Step 4: connect the measuring impedance and instrument, and select and record the bandwidth. Step 5: calibrate with the complete circuit de-energised by injecting a known charge at the object terminals. Step 6: remove the calibrator — never energise with the calibrator connected. Step 7: apply the voltage profile of the object's product standard. Step 8: record the PRPD for at least 60 s per level, plus inception and extinction voltage on a slow ramp. Step 9: inject artificial noise and repeat acquisition, applying filters, gating and cluster separation, comparing the diagnosis with and without cleaning. Step 10: final classification — defect type, level in pC, inception and extinction voltage, comparison against the normative limit, and technical recommendation. The closing exercise is a blind comparison: each pair receives another pair's PRPD with no context and must issue a diagnosis, which reveals whether the student reads the pattern or memorised the legend.

Key takeaway

A picocoulomb number with no source separation is not a diagnosis. Calibrate the complete circuit, document the bandwidth and every filter and gate you applied, separate the sources, then name each cluster's physical mechanism. And remember that the limit only means something inside the insulation technology it belongs to.

Module 18 of the Atlas Energy Academy · Diagnostic Methods · about 12 hours · Three-level series (Essentials, Pro, Masterclass) plus a live 3 h masterclass · Prerequisites: Part IV completed