The patterns are not a decoding table. They are signatures whose interpretation depends on the test arrangement, the applied voltage and the noise environment, and the same defect can produce different-looking patterns under different conditions. What the pattern does reliably is narrow the possibilities and identify the interference that should be excluded.

What a PD pattern represents

A discharge event occurs when the local field at a defect exceeds the breakdown strength of the small volume involved. Because the test voltage is alternating, the local field rises and falls through the cycle, so the defect discharges at particular points in the cycle rather than uniformly.

Plotting the events against phase produces clusters at those points. A defect that is physically symmetrical with respect to the field produces clusters at symmetric positions in the positive and negative half cycles. A defect that is not, such as a protrusion on one electrode, produces clusters with different amplitudes or positions between the two half cycles.

The number of events, their amplitude and their phase positions together form the signature. Reading it requires knowing the applied voltage at which it was recorded, because the pattern changes as the voltage rises: a defect that is inactive below its inception voltage produces nothing, and the pattern that appears at inception is not the same as the pattern at a higher test level.

Phase-resolved patterns and defect families

The defects that produce discharge fall into a small number of families, and each has a characteristic relationship between the applied voltage and the activity.

The families overlap, and the table describes tendencies rather than rules. Its practical use is to exclude possibilities: a pattern with sharply defined symmetric peaks is unlikely to come from a void deep in solid insulation, and a pattern that changes erratically between successive cycles is unlikely to come from a stable corona source.

Internal, surface and corona signatures

Internal discharge, from a void or a delamination inside solid insulation, is the family of most concern in a transformer or a cable because it erodes the insulation from within and its progression is difficult to arrest. Its signature is generally stable and reproducible, which is exactly what makes it worth monitoring: a pattern that repeats consistently and grows with voltage is evidence of a physical site rather than an artefact.

Surface discharge occurs along an interface between a solid insulator and a gas or a liquid. It tends to appear at lower voltages than internal discharge and to spread across a wider phase range, because the discharge follows the surface rather than a fixed cavity. Its practical significance depends on the material and on whether the surface is in a region of high tangential stress.

Corona at a sharp point or a loose conducting part produces the most recognisable signature: tightly clustered events at the voltage peaks, stable from the moment of inception and symmetrical between half cycles if the geometry is symmetrical. Corona in air outside the insulation system is often a maintenance finding rather than an insulation defect, which makes distinguishing it from internal activity particularly valuable.

Separating noise from discharge

Noise is the principal obstacle, and the substation environment supplies it continuously. The separation methods that work in practice are procedural, and the most useful is the relationship between the applied voltage and the observed activity.

Genuine discharge scales with the applied voltage: below inception there is nothing, and above it the amplitude and repetition rate grow in a repeatable way. Interference from radio sources and from communication equipment does not scale with the applied voltage at all. Corona from hardware in the substation scales with the system voltage rather than with the test voltage, so it changes only if the test arrangement changes.

The second method is comparison between the measured quantity and the phase position. Interference that couples into the measuring circuit tends to appear at a phase position determined by the interfering source rather than by the test voltage waveform, and it therefore appears at an arbitrary and unstable phase. Discharge from a physical site appears at a phase determined by the local field, and the position is stable between cycles.

The third method is a change of arrangement. Moving the coupling, changing the grounding point or taking a measurement with the test set energised but disconnected identifies which components of the signal come from the surroundings rather than from the test object.

Influence of test voltage and frequency

The test voltage determines what the measurement can find. A measurement taken at a voltage below the inception level of a defect will show nothing irrespective of the defect’s severity, which is why the applied voltage has to be stated with every pattern and why in-service measurements are limited in what they can conclude.

The inception and extinction voltages are the useful derived quantities. Inception is where activity first appears as the voltage rises; extinction is where it disappears as the voltage falls, and it is normally lower than inception. A defect whose inception voltage is close to the operating voltage is active whenever the equipment is in service, which is a stronger finding than a defect that only becomes active well above the operating level.

Test frequency changes the appearance of the pattern rather than the underlying physics. At the very low frequencies used on some cable tests, far fewer discharge events occur per cycle, so the display appears sparse and the classic phase-resolved clusters are formed from fewer points. The interpretation has to account for that, and an instrument configured for power frequency does not transfer directly to a low frequency test.

Recording patterns for comparison

The record is what makes a later measurement comparable. It should state the test voltage and the frequency, the coupling arrangement and the detection threshold achieved in that environment, the inception and extinction voltages where they were established, and the noise conditions during the measurement.

The pattern itself should be stored as raw phase-resolved data rather than only as an image. The same data can then be re-processed with different clustering or filtering parameters, which is often necessary when a first interpretation is questioned, and it can be compared numerically rather than visually with a later measurement.

Keeping the calibration record with the data matters for the same reason. The charge scale factor applied to the measurement determines what the vertical axis of the pattern means, and a pattern from which the scale factor has been lost cannot be compared with a later one in absolute terms.

When pattern recognition is not conclusive

Pattern recognition is inconclusive more often than the literature suggests, and the situations have a common structure. The first is a low event count, where too few events are recorded to form a reliable distribution. That happens at low test frequencies, on small test objects and where the detection threshold is high relative to the activity.

The second is mixed activity, where two or more sources are active simultaneously and the resulting pattern is a superposition that matches no single family. The third is a pattern dominated by interference, where the underlying defect signature is present but buried in the noise.

In each case the correct response is to change the measurement rather than to force an interpretation. Raising the test voltage where the standard allows, changing the coupling, using an acoustic or UHF measurement to locate the source, or repeating the measurement in a quieter period are all steps that resolve the question better than analysis of an ambiguous pattern.

Reporting language that avoids overclaiming

A discharge report has to state what the measurement supports. A pattern consistent with internal discharge in solid insulation, recorded at a stated test voltage, with a stated detection threshold and a stated inception voltage, is a defensible statement. A conclusion that the insulation contains a void is not, because the pattern narrows the family rather than identifying the site.

The report should separate the measurement from the interpretation. The measurement section carries the applied voltage, the magnitude, the threshold and the raw pattern. The interpretation section names the defect families the pattern is consistent with, the evidence that excludes the others, and the further tests that would resolve the remaining ambiguity.

Where the measurement supports a decision, the report should state the decision and the evidence behind it. A recommendation to reduce the monitoring interval is supported by a reproducible pattern that grows with voltage and by an inception voltage near the operating level. The same recommendation based on a single pattern recorded at one voltage is not, and a report that presents the two as equivalent will eventually be tested by a failure that the evidence did not support. The measurement method and its calibration requirements are defined in IEC 60270, the field test framework that governs the applied voltage is published as IEC 60060-3, instrument method documentation is published by suppliers such as OMICRON, and the accumulated field experience with discharge diagnostics is coordinated through CIGRE study committees and published in the transmission research programme of EPRI.