Insulation testing in a wind turbine is not the same job as insulation testing in a plant motor room. The machine sits 80 to 150 meters up, inside a steel box that breathes damp air all night. It spends half its life stopped and cold. It is bolted to a power converter that a 500 V megohmmeter will destroy in one shot. And the crew that tested it last year is probably not the crew testing it today.
The measurement itself is simple. Everything around it is where the results get lost.
Table of Contents
Part 1 — The asset list
“Testing the turbine” usually means testing five separate things. They have different voltage ratings, different acceptance limits, and different isolation problems.
| Asset | Typical rating | Test voltage | Watch for |
|---|---|---|---|
| Generator stator (DFIG) | 690 V | 500 V DC | Stator-side converter, RTDs, heaters |
| Generator stator (MV PMG / geared) | 3.3 kV or 6.6 kV | 1000–2500 V DC | Stress grading on endwindings |
| DFIG rotor winding | 690 V class, slip-dependent | 500 V DC | Brush gear lifted, carbon dust on rings |
| Tower drop cable | Matches generator | Same as generator | Test it separately at least once |
| Yaw and pitch motors | 400–690 V | 500 V DC | Small random-wound, PI often useless |
| Nacelle or base transformer | Varies | Per transformer procedure | Separate procedure, not covered here |
Two notes on the edges of that list. IEEE 43 is written for rotating machines rated 750 W and above and does not cover fractional-horsepower machines, so the smallest pitch motors fall outside it. And blade heating elements in cold-climate turbines are resistive loads, not windings — treat them as a separate insulation check against the blade structure, not as a polarization index candidate.
Part 2 — Isolation, before anything else
This is the part that wrecks equipment.
A modern wind generator is never a standalone winding. It is a winding wired directly into a power electronic converter. Applying 500 V DC through the machine terminals with the converter still connected puts that voltage across IGBT modules and DC-link capacitors that are rated for nothing of the sort. The fault is instant and expensive, and it looks nothing like an insulation problem afterward.
Disconnect and ground before the leads go on:
- Stator-side and rotor-side converter connections
- DC link and any dv/dt or sine filter
- RTDs and PT100 elements in the slots and bearings
- Anti-condensation heaters
- Encoder and speed sensor wiring
- Surge arresters and any capacitor bank at the terminals
- Slip ring brush gear, lifted clear of the rings
Whatever stays connected changes the reading. If you cannot practically remove something — a short section of terminal cable, for instance — leave it connected, but write it down. A trend built on “generator only” one year and “generator plus 90 m of tower cable” the next year is not a trend. It is two unrelated numbers.
The same logic applies to test point. Testing from the tower base is faster and safer than climbing. It is also a different measurement. Pick one and stay with it, or take both and label them.
Part 3 — The cold nacelle problem
Insulation resistance falls as temperature rises, on an exponential curve. Correcting every reading to a 40 °C base is the only way two tests taken a year apart mean anything.
Wind generator windings are thermosetting — epoxy or polyester systems. Use the thermosetting correction factors, not the older asphaltic ones. The difference is large:
| Winding temperature | Thermosetting K<sub>T</sub> | Thermoplastic K<sub>T</sub> |
|---|---|---|
| 10 °C | 0.7 | 0.125 |
| 20 °C | 0.8 | 0.25 |
| 30 °C | 0.9 | 0.5 |
| 40 °C | 1.0 | 1 |
| 50 °C | 1.5 | 2 |
| 60 °C | 2.3 | 4 |
| 70 °C | 3.3 | 8 |
| 80 °C | 4.6 | 16 |
Corrected resistance is the measured resistance multiplied by K<sub>T</sub>.
Two cautions that matter more in wind than anywhere else.
The correction has a validated range. The published equations are approximations, and they can produce significant errors outside roughly 10 °C to 60 °C. A nacelle in a northern winter can hold a stopped generator at 0 °C or below. At that point you are extrapolating. Record the raw value and the temperature, note that correction was out of range, and lean on history from similar conditions rather than on a corrected number you cannot defend.
Use the winding temperature, not the nacelle temperature. The embedded RTDs are there. Read them. A generator that ran until this morning can be 40 K above the air around it, and it will keep dropping through a ten-minute test.
That last point has a side effect on polarization index. If the winding is hot at the start, it cools during the test, resistance rises for thermal reasons, and the PI comes out artificially high. When a machine reads a suspiciously good PI straight off a hot stop, repeat it at or below 40 °C.
Part 4 — Moisture and the dew point
If the winding surface is at or below the dew point of the nacelle air, a moisture film forms on the insulation. Resistance drops, PI drops, and neither number tells you anything about the insulation wall.
The practical rules:
- Test a machine that has been running before it cools below the dew point.
- If the turbine has been stopped and cold, energize the anti-condensation heaters and let the winding come up first.
- Record ambient temperature, relative humidity, dew point, winding temperature, and how long the turbine has been out of service. All five.
- Absorbed moisture drives PI toward 1. A low PI on a cold, damp, long-idle turbine is a moisture reading until proven otherwise.
Contamination compounds it. Dust and salts that are harmless when dry become partly conductive when wet. Offshore and coastal sites see this constantly. Surface contamination usually cleans up — insulation resistance and PI recover after cleaning and drying, which is itself a useful diagnostic. If cleaning restores the number, the wall was never the problem.
Part 5 — Test voltage
Match the DC test voltage to the winding rating, using rated line-to-line voltage for three-phase machines:
| Winding rated voltage | DC test voltage |
|---|---|
| Below 1000 V | 500 V |
| 1000–2500 V | 500–1000 V |
| 2501–5000 V | 1000–2500 V |
| 5001–12 000 V | 2500–5000 V |
| Above 12 000 V | 5000–10 000 V |
A 690 V DFIG stator gets 500 V. Not 1000 V because the tester has the range and it “gives a cleaner number.” Overstressing a wet or aged low-voltage winding is a real way to turn a test into a failure.
Test at negative polarity. The reason is electroendosmosis — on older or wet windings, reversing lead polarity can produce a noticeably different resistance, and negative polarity is the convention that keeps results comparable. Most modern testers do this by default; confirm rather than assume.
One more voltage effect worth knowing. On a dry, healthy winding, resistance stays broadly the same across test voltages up to the peak of rated voltage. A resistance that drops sharply as you step the voltage up points at fractures, concentrated dirt, or moisture — not at a normal winding.
Part 6 — Acceptance values, and the trap in them
Polarization index minimums, by thermal class of the insulation:
| Thermal class | Minimum PI |
|---|---|
| Class 105 (A) | 1.5 |
| Class 130 (B) and above | 2.0 |
Wind generators are Class F or H. The number to beat is 2.0.
One-minute insulation resistance minimums, corrected to 40 °C, for the whole winding:
| Machine type | Minimum IR₁ |
|---|---|
| Random-wound stators, and form-wound coils rated below 1 kV | 5 MΩ |
| Most AC form-wound windings built after about 1970 | 100 MΩ |
| Pre-1970 windings, all field windings, and anything not covered above | kV + 1 |
A 690 V DFIG stator therefore has a floor of 5 MΩ. That floor is close to meaningless as a health indicator. A generator reading 8 MΩ passes and is in serious trouble. The floor is a return-to-service gate, not a condition assessment. Condition comes from the trend.
Now the trap. When the corrected one-minute resistance exceeds 5000 MΩ, polarization index stops being reliable and is not recommended as an assessment tool. At that resistance the total current is down in the sub-microamp range, and supply ripple, humidity, lead placement, and connection quality move the reading as much as the insulation does.
This is the single most common misread in wind O&M. A dry, healthy, well-impregnated 690 V generator in a warm nacelle reads 12 GΩ at one minute and a PI of 1.1. The crew flags it. There is nothing wrong with it. Above 5000 MΩ, report the resistance and drop the PI.
A second inflation source, on MV machines only: continuous stress control material over the endwinding overhang. Where stress grading tape contacts bare copper at the bar ends, surface leakage can dominate, and the PI can be pushed far from what the bulk insulation would give — measured differences of a factor of two or more against the same bar with a different grading system. If your MV generator has SiC-based grading, treat an unusually high PI with the same suspicion as an unusually low one, and use a guard connection to strip the surface path out of the measurement.
Part 7 — The DFIG rotor
The rotor is the part crews skip, and it fails as often as the stator.
The rotor winding on a doubly-fed machine is fully insulated and encapsulated, so polarization index applies to it — unlike a squirrel-cage rotor, where the bars are not insulated from the body and PI has no meaning at all.
Access is through the slip rings. That means:
- Lift the brushes clear before testing. Carbon dust bridges rings and gives you a leakage path that has nothing to do with the winding.
- Clean the rings and the surrounding housing first. Slip ring chambers accumulate conductive dust faster than any other part of the machine.
- Isolate the rotor-side converter, same as the stator.
- Read across the rings to the shaft or frame, and record which reference you used.
A rotor circuit whose resistance drops after a brush change and recovers after a chamber clean is telling you about housekeeping, not insulation. That distinction is worth writing on the sheet.
Part 8 — Discharge, and why it matters at 100 meters
The winding stores charge. Removing the lead does not remove it.
Ground for at least four times the voltage application time. A ten-minute PI test means a forty-minute ground. Long tower cables add capacitance and make it worse. The test is not finished when the reading is written down — it is finished when the winding is discharged and no measurable return voltage remains.
Two consequences.
For safety: nobody handles a lead or a terminal until discharge current is essentially zero and there is no discernible return voltage after the ground is removed. In a nacelle, that means the discharge window has to be built into the job plan alongside rotor lock, yaw lock, and fall protection. The temptation to unclip and climb down is exactly when people get hit.
For accuracy: residual charge or unrelaxed polarization in the insulation corrupts the next measurement. If you are testing three phases in sequence, the second and third readings are wrong unless each winding was fully discharged first. Measure discharge current at the start of each test as a check that you actually got there.
Part 9 — What this test will not find
DC insulation resistance and polarization index detect contamination, absorbed moisture, and severe cracking. That is the whole list.
They will not tell you:
- Dielectric strength. There is no resistance value at which a winding is about to fail. The two are not directly related unless a defect is concentrated.
- Internal voids. Poor impregnation, thermal deterioration, and thermal cycling damage in form-wound coils can be invisible to a DC test. One layer of intact mica tape blocks the current path, so the voids behind it never show up. AC-based tests — capacitance, dissipation factor, partial discharge — are far more sensitive to internal problems.
- Anything that only happens while turning. Standstill testing cannot see loose coils, endwinding movement, or vibration-driven wear. In a machine that thermal-cycles as hard as a wind generator, that is a large blind spot.
- Turn-to-turn insulation. A megohmmeter measures to ground. Turn insulation needs a surge comparison test.
So an insulation resistance test that passes is not a clean bill of health. It is a gate: the winding is dry and clean enough to energize, and dry and clean enough to be worth a higher-voltage or AC diagnostic test. That second use matters. A machine with low PI and low IR should not be pushed into further high-voltage testing — you clean and dry it first, or you find out the hard way.
Part 10 — A record sheet that survives crew turnover
Trending only works if the conditions are comparable, and comparable means recorded. Wind sites have high technician turnover and multi-vendor service contracts, so the sheet has to carry everything the next person needs.
Minimum fields per test:
| Field | Why it is there |
|---|---|
| Turbine ID and asset tested | Stator, rotor, cable, or combination |
| Date, time, hours since shutdown | Thermal and moisture state |
| Winding temperature and source | RTD tag, not nacelle air |
| Ambient temperature, RH, dew point | Condensation risk assessment |
| Test voltage and polarity | Comparability |
| Test point | Nacelle terminals or tower base |
| What remained connected | The single biggest source of year-to-year error |
| IR at 30 s, 1 min, 10 min | Enables PI and shorter-ratio variants |
| Correction factor applied and corrected IR | With a flag if temperature was outside 10–60 °C |
| Instrument model and lead configuration | Guard used or not |
On that ninth row: for modern windings, absorption current can decay to near zero in two or three minutes, which is why some organizations use shorter ratios — one minute over thirty seconds, or five minutes over one minute. They cut climb time substantially and lose little information about contamination or moisture. The honest caveat is that no standardized pass-fail criteria exist for those variants. Record the full ten-minute curve where you can, so the data stays useful if your fleet later adopts a shorter ratio.
FAQ
What test voltage for a 690 V wind generator? 500 V DC. The winding is rated below 1000 V, which puts it in the lowest band. Higher voltages risk overstressing a wet or aged low-voltage winding.
My PI is 1.2 but the resistance is 10 GΩ. Is the generator bad? Almost certainly not. Above about 5000 MΩ corrected, polarization index is not a reliable assessment tool — the currents involved are too small and too easily disturbed. Report the resistance, note the PI as not applicable, and compare against previous resistance readings taken under similar conditions.
Can I test through the converter? No. Disconnect the converter, the DC link, and any filters before applying test voltage. DC insulation test voltage across power electronics destroys them.
Does polarization index apply to the rotor? On a doubly-fed machine, yes — the rotor winding is insulated and encapsulated. On a squirrel-cage rotor, no. The bars are not insulated from the rotor body, so there is no insulation system to polarize.
How long do I have to keep the winding grounded afterward? At least four times the duration of the voltage application. A ten-minute test needs forty minutes of grounding. Confirm with a return voltage check before anyone touches the terminals.
Winding is at 5 °C. Do I still correct to 40 °C? Correct it, but flag it. The published correction equations are approximations with a validated range of roughly 10 °C to 60 °C, and below that they can introduce significant error. Weight the comparison toward previous tests taken under similar cold conditions rather than toward the corrected value alone.
Should I test the tower cable with the generator or separately? Separately at least once, so you have a baseline for each. Then pick one method for routine testing and never change it mid-trend. A cable insulation problem and a generator insulation problem look identical from the tower base.
IR and PI both pass. Is the winding healthy? It is dry and clean. That is all these tests establish. Internal voids, thermal aging, turn insulation faults, and vibration damage need capacitance and dissipation factor, partial discharge, or surge comparison testing to detect.