Generator and Large Motor Insulation Testing

In machines, trend and pattern outrank any absolute number.

Rotating machines close this part of the track with a provocation that subverts everything before it: in generators and large motors, some partial discharge is acceptable. Mica-epoxy insulation lives for decades with PD levels that would condemn a transformer in minutes of testing. That tolerance is not regulatory sloppiness. It is a direct consequence of material physics: mica, an inorganic barrier, resists discharge erosion in a way no impregnated paper and no polyethylene can.

The price of that robustness is a diagnostic discipline of its own, with its own standards, its own criteria — trend and pattern, never zero picocoulombs — and an unusual integration between electrical and mechanical evidence. The most destructive dielectric defect in a stator, slot discharge, begins with a loose wedge. A machine moves: stator bars vibrate under electromagnetic force proportional to the square of current, dominated by twice line frequency. Its insulation is a heterogeneous composite full of interfaces where microcavities are unavoidable from manufacture. And its dielectric life is too long, 30 to 50 years, for perfection to be demanded. What is demanded instead is stability.

What you will be able to do

• Explain why mica-epoxy Type II systems tolerate partial discharge in operation and why machine assessment is always by trend and pattern, never by isolated absolute amplitude.

• Run and interpret insulation resistance and polarisation index per IEEE 43 and IEC 60034-27-4, applying temperature correction to 40 °C and recognising the minimums by construction family: 100 MΩ for modern form-wound, kV + 1 MΩ for pre-1970 windings, 5 MΩ for random-wound.

• Specify the routine applied voltage of 2·Un + 1 kV for one minute and judge the field alternatives — DC at 1.7× and VLF per IEEE 433-2022 — with their advantages, costs and equivalence limits.

• Conduct a surge test per IEEE 522-2023, comparing waveforms between phases and interpreting the error area ratio — the only test that sees a turn-to-turn short before it becomes a failure.

• Take open-circuit and short-circuit characteristics of synchronous generators per IEEE 115-2019, extracting reactances, short-circuit ratio and the limits of the capability curve.

• Interpret tan delta and tip-up per IEC 60034-27-3, off-line and on-line PD maps per IEC 60034-27-1 and -27-2, core testing, wedge maps and rotor tests, integrating everything into a single verdict: operate, monitor or rewind.

Trend is the master criterion

The groundwall of a high voltage form-wound bar or coil is built from successive layers of mica tape over a transposed Roebel conductor, consolidated with epoxy resin. Each component has a defined job: mica laminae form the inorganic barrier that resists discharge erosion; the resin fills voids, bonds layers and transfers mechanical stress; the semiconductive slot coating maintains equipotential contact between the bar surface and the grounded core; silicon carbide stress grading at the slot exit relieves the field in the transition to the end winding; and wedges, side packing and blocking immobilise the assembly against electromagnetic force.

Deterioration rarely starts in the mica. It starts in the resin and at the interfaces — internal micropores, mica-resin delamination, gaps between bar and core, surface contamination, the slot exit region and the end winding. The PD born in those sites erodes the organic component first, and mica barriers slow the progression to failure by years. That is precisely why mica-based systems are classified as Type II, designed to operate with some PD activity through life, in contrast with Type I polymeric systems, typically low voltage converter-fed machines, where any PD is degenerative.

Insulation resistance and polarisation index: 40 °C or nothing

When DC voltage is applied to a winding, the total current is the sum of five components: capacitive charging current, dominant in the first seconds; absorption or polarisation current in the resin, decaying over minutes; volume conduction through the insulator; surface leakage, sensitive to contamination and moisture; and current drained by the semiconductive and grading coatings. Apparent resistance R(t) = V_DC/I_T(t) rises with time as the first two components decay, and that absorption dynamic is what the polarisation index captures. IEC 60034-27-4 warns that moisture, surface contamination, grading coatings and water cooling circuits can dominate the measurement and completely mask the state of the groundwall.

The minimums by construction family matter. Modern form-wound thermoset mica-epoxy windings at or above 1 kV: 100 MΩ at 40 °C. Windings made before roughly 1970 with shellac or asphalt: kV + 1 MΩ, where kV is rated line-to-line voltage. Random-wound and form-wound below 1 kV, and DC armatures: 5 MΩ. A worked example is obligatory: for a 6.6 kV modern form-wound motor, the reference minimum is 100 MΩ, not the 7.6 MΩ the historic kV + 1 rule would suggest. And meeting the minimum does not certify health. A machine that historically read 20 GΩ and now reads 150 MΩ has moved two orders of magnitude — passed by the absolute minimum and failed by trend.

Three waveshapes for the groundwall, and one for the turns

The base rule of IEC 60034-1 for new machine windings is a test voltage of 2·Un + 1 kV, RMS, at power frequency, for 60 seconds, applied between winding and ground with the other windings and the core grounded. For 6.6 kV that is 14.2 kV; for 13.8 kV, 28.6 kV. The acceptance criterion is binary: no breakdown, no flashover, no voltage collapse and no fault current protection operation during the minute. The source has to supply the capacitive current of the winding, and large hydro generators are loads of hundreds of nanofarads to microfarads, which at 50 or 60 Hz means considerable reactive power. That is exactly why resonant sources and VLF exist.

None of those three reaches the defect that brings down most motors: the turn-to-turn short. Insulation between adjacent turns of a form-wound coil is thin, designed for volts per turn in steady state, but exposed to thousands of volts when a fast front — a vacuum breaker operation, converter switching, a transferred lightning surge — distributes non-uniformly through the winding, with the first turns of the first coil taking most of the front. The surge test reproduces exactly that stress. A charged capacitor discharges into the winding, forming an LC circuit whose damped oscillation depends on winding inductance. Two identical phases produce identical waveforms; a turn short reduces the inductance of the affected phase, raising oscillation frequency and damping. The historic level for a new form-wound winding starts from V1 = 0.816·Un per unit, giving V_surge = 3.5·V1 ≈ 2.86·Un peak — 18.9 kV peak for 6.6 kV, 39.4 kV peak for 13.8 kV. Front time matters as much as amplitude: fronts of 0.1 to 0.2 µs concentrate the stress between turns, while fronts above about 1.2 µs transfer it mostly to the groundwall and stop testing what you meant to test.

Slot discharge: a mechanical defect wearing a dielectric mask

The causal chain at the heart of this module runs as follows. The wedge loses tightness. The bar begins to vibrate under electromagnetic force at twice line frequency, 100 or 120 Hz. The vibration abrades the semiconductive coating against the core laminations. With equipotential contact lost, the bar surface floats in potential relative to the grounded core. Slot discharge establishes itself as high-energy PD across the gap. Erosion advances from the surface into the groundwall until phase-to-ground failure.

The wedge tap map turns percussion into engineering data: every slot and axial position is numbered, a standardised impact is applied, and the acoustic response is classified as tight, intermediate or loose, with cracked or missing wedges recorded. Electronic systems measure frequency, duration and energy of the response, reducing subjectivity. Typical field criteria, and there is no universal IEC limit here: a missing or broken wedge is unacceptable; 20 to 25% loose wedges in one slot, or loose wedges contiguous at the ends where bar force is highest, triggers rewedging.

There is no universal pC limit — and that is the point

Tan delta measured on individual bars or by phase, in voltage steps of 0.2·Un, turns dielectric loss into a measure of internal PD activity. Below the inception voltage, tan delta grows slowly with voltage; above inception, each step adds discharge energy to the losses, and that growth is the tip-up. IEC 60034-27-3 limits for new bars and coils up to 21 kV with guard rings: tan delta at 0.2·Un no greater than 20 × 10⁻³; increment per 0.2·Un step no greater than 5 × 10⁻³; and tip-up between 0.6·Un and 0.2·Un no greater than 5 × 10⁻³. Interpretation: a high initial tan delta with low tip-up suggests incomplete cure or conductive contamination, while high tip-up with normal initial value points to internal voids and delamination. In complete windings the reading is diluted, since hundreds of parallel bars mask one bad bar, which is why per-bar testing in the factory is irreplaceable. Capacitance should always be recorded alongside tan delta, because a change in capacitance between campaigns indicates delamination or ingress of moisture.

On-line PD measurement under real voltage and load uses permanent capacitive couplers at the terminals, high-frequency current transformers at the neutral, or slot sensors. It is the only method that sees the effect of load through bar vibration, of temperature and of excitation, and it is the basis of continuous monitoring. Its central challenge is noise: static excitation, brushes, converters and external corona all produce pulses that imitate PD. Separation relies on phase synchronism, comparison between couplers using arrival time and direction, pulse-shape clustering, and correlation with operating conditions. Pulses independent of voltage phase, fixed lines in the pattern, and identical signatures on all phases indicate electronic noise, not PD.

Syllabus

• The mica-epoxy groundwall: layers, semiconductive slot coating, stress grading, and the mechanical wedge system

• Type I versus Type II insulation systems and what that classification licenses

• Global VPI versus resin-rich manufacture and the different defect patterns of each

• Thermal classes, the class F insulation with class B rise practice, and the Montsinger approximation

• Standards status: which IEEE documents are inactive-reserved and what replaced them

• The five currents in an insulation resistance measurement

• Temperature correction to 40 °C: R40 = RT · 0.5^((40−T)/X) and why the ritual is non-negotiable

• Polarisation index, its class minimums, and when IEEE 43 declares it meaningless

• Routine applied voltage 2·Un + 1 kV: levels by rated voltage and the capacitive current problem

• Field alternatives compared: AC 50/60 Hz, DC at 1.7×, VLF at 0.1 Hz

• Surge test: front time, level at about 2.86·Un peak, error area ratio bands, disciplined procedure

• Open-circuit and short-circuit characteristics, reactances, short-circuit ratio, capability curve boundaries

• Temperature rise, efficiency and loss segregation, overspeed, vibration and noise

• Stator diagnostics: tan delta and tip-up, off-line and on-line PD, PRPD patterns, core testing, wedge maps

• Rotor diagnostics: recurrent surge oscillography, pole voltage drop, air gap, shaft currents and bearing insulation

• The slot discharge causal chain from loose wedge to ground fault

Laboratory work

A case-study exercise built on three complete result sets from real machines — IR and PI, tan delta and tip-up, PD map, wedge map and rotor tests — requiring an engineering opinion at three levels: operate (return to service with normal monitoring), monitor (return with intensified monitoring and scheduled reinspection), or rewind (intervene before return). The debriefing forces a hierarchy of evidence. In one machine, no single number fails: IR and PI are excellent, but the convergence of rising PD trend, a slot pattern and contiguous loose wedges tells one coherent story — the causal chain in progress. In another, an IR of 8 MΩ against a 100 MΩ minimum suspends any high voltage test immediately; drying is attempted, and a surge test run after partial IR recovery seals the diagnosis of an incipient turn-to-turn short.

Key takeaway

Correct to 40 °C or do not compare at all. Use the surge test for turn insulation, because a phase-to-ground hipot cannot see the defect that fails most motors. And before you interpret a PRPD, read the wedge map — slot discharge is mechanical at the origin, and the wedge tells you first.

Module 17 of the Atlas Energy Academy · Equipment Families · about 6 hours · Video series (10 × 12–18 min) plus case-study practice · Prerequisites: Module 11 — Test Taxonomy