A withstand test tells you the insulation did not break down for one minute. It tells you nothing about what happens over ten years. Solid insulation with a void inside will pass a hipot shot and still erode from the inside out, one small discharge at a time, until it doesn’t.
That’s the gap a partial discharge test fills. You’re not asking whether the material survives a voltage. You’re asking whether it discharges at all while it sits at the voltage it will actually live at.
For low-voltage equipment, the framework is IEC 60664-1, with the measurement method itself following IEC 60270. The test is generally applied to components, small assemblies, and small equipment rather than to a full panel.
Table of Contents
Why solid insulation gets its own test
Air recovers. A clearance that flashes over is undamaged once the arc goes out. Solid insulation does not recover. Every discharge inside a void removes a little material, and the cavity grows.
That’s why a design can be dimensionally correct and still fail. Your clearance and creepage distances can be right on the drawing while the material bridging them has a void from a bad molding shot. The dimensions are checked with a ruler. The material has to be checked electrically.
Two circuits, and why one is usually better
Both test circuits do the same job. A test voltage feeds the specimen. A coupling capacitor sits in parallel. A measuring impedance in series with the coupling branch picks up the discharge pulses and feeds the PD meter. An optional filter keeps charge from bleeding back into the supply.
The difference is where the specimen sits relative to earth.
Earthed specimen. One terminal of the specimen goes to earth, and the measuring impedance sits in the coupling capacitor branch.
Unearthed specimen. The specimen floats, and the measuring impedance sits in the specimen’s return leg.
On paper they’re equivalent. In practice the stray capacitance at the high-voltage terminal works against you in the earthed arrangement and works for you in the unearthed one. That makes the unearthed circuit the better default for sensitivity.
Component notes that matter:
- The coupling capacitor must be low inductance, must be discharge-free up to the highest test voltage you’ll use, and needs a resonant frequency well above the measuring band. A coupling capacitor that discharges is just a noise source in series with your measurement.
- The measuring impedance should drop almost nothing at test frequency, while still giving usable sensitivity at the measuring frequency.
- The filter is optional. If you use one, it needs to be a high impedance at the measuring frequency.
For sensitivity, you want the coupling capacitor larger than the specimen plus its stray capacitance, a high filter impedance, and a low measuring impedance. A high-capacitance specimen fights all of this, which is why big objects are hard to measure at low charge levels.
What has to be settled before you energize
Four things need a number before the test means anything:
Test frequency. Normally power frequency, AC. Harmonic distortion needs to stay under 3 percent, which keeps the peak value predictable from an RMS reading. DC is generally a poor fit here, partly because a noise-free DC environment is hard to achieve and partly because the voltage distributes differently.
Specified discharge magnitude. The pass/fail charge level. The guidance is to set the lowest practical value. Above 10 pC the number stops being useful for most equipment, and modern instruments can resolve down around 2 pC.
Climatic conditions. Room temperature and mid-range humidity, roughly 23 °C and 50 percent RH.
Measuring band. The lower measuring frequency should sit at least ten times above the test frequency, which keeps the power frequency itself out of the reading. The upper end doesn’t need to go past 2 MHz. Narrowband and wideband instruments both work. Radio interference meters can only be used with their weighting filter disconnected, because that filter is built to model how noise sounds to a human ear, not to measure charge.
The voltage profile: raise high, then drop
This is the part that trips people up. The test voltage does not sit at one level.
The reason is hysteresis. The voltage that starts a discharge (inception) is higher than the voltage that stops one (extinction). In service, a transient can kick discharges into life, and then a much lower steady voltage keeps them going. Testing at the steady voltage alone would never start them, so you’d pass a part that fails in the field.
The workaround is to start high enough to initiate discharges, then drop to the level you actually care about and see if they persist.
The multipliers stack onto the recurring peak voltage:
| Factor | Value | What it covers |
|---|---|---|
| F1 | 1.2 | Basic safety margin; environmental effects on extinction voltage |
| F2 | 1.25 | Inception-to-extinction hysteresis |
| F3 | 1.25 | Extra margin for reinforced insulation |
| F4 | 1.1 | Mains supply deviation from nominal |
For basic and supplementary insulation, the starting voltage works out to 1.5 times the recurring peak. For reinforced insulation, F3 pushes it to 1.875 times.
The sequence:
- Raise uniformly from zero to the initial (elevated) voltage.
- Hold it for a short window, no more than about 5 s.
- If nothing discharges, drop to zero. Done.
- If a discharge occurs, drop to the actual test voltage and hold it while you measure the magnitude.
The part that passes is the part where discharges either never appeared, or died back below the specified magnitude once the voltage came down.
Calibration and the noise floor
A charge reading in pC is meaningless until the circuit is calibrated, because charge is lost into the source and into stray capacitance before it ever reaches your meter. Calibration accounts for those losses.
Calibrate first without the specimen. Swap in a discharge-free capacitor of similar impedance. Short the transformer primary and leave it de-energized. Inject a known charge at the specified magnitude and set the meter to read it.
Then check your noise floor. With that same arrangement, raise the voltage to the highest level you’ll use and watch the noise. If noise exceeds half the specified discharge magnitude, the test is not valid yet — fix the noise before going further.
Then calibrate again with the real specimen in circuit. Any change to the circuit or the specimen means recalibrating. Across a run of similar parts you can calibrate occasionally instead of every time, but only if the coupling capacitor impedance is well below the specimen’s, or the specimen impedance stays within about ±10 percent of the calibrated value.
Skipping recalibration is the quiet failure mode here. A circuit that has lost sensitivity does not report an error. It just reports low numbers, and harmful discharges go undetected.
Getting the noise down
Expect trouble in an unscreened industrial bay. Single noise pulses can reach 100 pC there, and even in decent conditions 20 pC is a realistic floor. Getting to 1 pC takes screening, disciplined earthing, and filtering on the supply input.
Noise comes from two places.
With the circuit de-energized — switching in nearby circuits, coupling in conductively through the mains connection or electromagnetically through the air. The electromagnetic path is still there with the supply switched off.
With the circuit energized — discharges from everything that isn’t your specimen. The test transformer, the HV leads, bushings, and any poor contact. This kind of noise climbs with test voltage, which makes it easy to mistake for the specimen breaking down.
What works:
- Line filters on the feed for conductively coupled noise.
- No earth loops.
- For narrowband meters, shift the measuring frequency away from the interfering signal. This is the cheapest fix available and it solves most radio-band problems.
- For wideband meters, band-stop filters help against narrow interferers. Broadband noise only responds to screening.
- A fully enclosed, high-conductivity screen is the strongest measure.
Before blaming a part, disconnect it and re-run at voltage. If the noise stays, it was never the specimen.
Reading the result
The insulation passes when no breakdown occurred, and either no discharges appeared at all, or the magnitude after the voltage was lowered stayed at or under the specified value.
One rule that catches people: you do not subtract the noise floor from the meter reading. The number on the display is the number you judge. That’s exactly why the noise floor has to sit under half the specified magnitude before you start — the margin has to be built into the setup, not into the arithmetic afterward.
Measurement accuracy for discharge magnitude is ±10 percent or ±1 pC, whichever is larger. At single-digit pC targets, the ±1 pC floor is what governs.
Test setup reference
| Parameter | Requirement |
|---|---|
| Test voltage type | AC, normally power frequency |
| Harmonic distortion | Under 3 percent |
| Preferred circuit | Unearthed specimen (better sensitivity) |
| Measuring band, lower limit | At least 10 × test frequency |
| Measuring band, upper limit | No need to exceed 2 MHz |
| Pulse resolution time | Under 100 µs |
| Climatic conditions | About 23 °C, 50 percent RH |
| Practical magnitude ceiling | Values above 10 pC are not useful |
| Achievable magnitude | Down to about 2 pC |
| Maximum noise level | Under 50 percent of specified magnitude |
| Noise handling | Not subtracted from the reading |
| Magnitude accuracy | ±10 percent or ±1 pC, whichever is greater |
| Initial voltage, basic insulation | 1.5 × recurring peak |
| Initial voltage, reinforced insulation | 1.875 × recurring peak |
| Initial hold time | 5 s maximum |
FAQ
Why start above the voltage you’re actually testing for?
Because discharges start at a higher voltage than they stop at. In service a transient starts them and normal voltage sustains them. The elevated start reproduces that trigger, then the drop shows whether they persist.
Can I subtract background noise from the measured charge?
No. The reading stands as measured. Get the noise below half the specified magnitude first, then test.
Earthed or unearthed specimen circuit?
Unearthed, unless something about the specimen forces otherwise. Stray capacitance at the HV terminal degrades sensitivity in the earthed circuit and improves it in the unearthed one.
Why does a large specimen give worse sensitivity?
More of the discharge charge is lost into the specimen’s own capacitance and into stray paths before reaching the measuring impedance. Calibration corrects the scale but cannot recover what never arrived.
How often does the circuit need recalibration?
Any time the circuit or the specimen changes. For a batch of similar parts, occasional recalibration works if the coupling capacitor impedance stays well below the specimen’s, or the specimen impedance holds within about ±10 percent of the calibrated condition.
Does a partial discharge test replace the withstand test?
No. They answer different questions. A withstand test asks whether the insulation survives a defined overvoltage. A PD test asks whether it degrades quietly at normal working voltage.