AC High Voltage Generation

The capacitive load decides the topology, long before the voltage does.

Everything starts at the test transformer. The power-frequency withstand test — the classic applied voltage test — is the most executed test in any high voltage laboratory in the world. It certifies insulators, bushings, instrument transformers, arresters, cables and protective equipment. Its apparent simplicity, raise the voltage, hold one minute, watch, hides a chain of engineering decisions that begins long before energisation: which source topology, how much installed power, and how to guarantee that the waveshape delivered to the object is the one the standard recognises as valid.

The central difficulty is that the test object is almost never resistive. Insulators, bushings, cables and windings look like capacitors to the source, from tens of picofarads to microfarads. The current the source must deliver is predominantly reactive and grows linearly with frequency, capacitance and voltage; apparent power grows with the square of voltage. That physics is what separates testing one insulator, a few hundred volt-amperes, from testing a kilometre of extruded cable, megavolt-amperes. It is also what organises the three topologies in this module.

What you will be able to do

• Size a test source from the capacitive load using S ≈ ω·C·U² and P ≈ k·ω·C·U², and interpret the result against the power actually available in the laboratory.

• Compare the single test transformer, the cascade and the series resonant circuit on principle, advantages, technical limits and cost, and select the correct topology for a given object.

• Operate the voltage control — variable autotransformer, column regulator or static converter — and explain the sinusoidal waveshape requirement: peak/RMS equal to √2 within ±5%.

• State the IEC 60060-1 requirements for AC test voltage: test value defined as peak divided by √2, frequency band 45 to 65 Hz for standard power-frequency tests, level tolerance ±1% for tests up to 60 s, and a ramp of about 2% of U per second above 75% of the test value.

• Explain why the internal series impedance of a test transformer raises rather than lowers the terminal voltage under capacitive load, and why measurement must always sit on the high voltage side.

• Recognise accidental series resonance between the source leakage inductance and the object capacitance, and calculate f₀ = 1/(2π√(LC)) before energising.

A test transformer is a distant relative of a power transformer

The resemblance ends at electromagnetic induction. A substation power transformer runs continuously with a moderate ratio and hundreds of amperes on the secondary. A test transformer steps 220 V or 380 V directly to 100 to 600 kV, ratios of 500:1 to 2000:1, delivering hundreds of milliamperes to a few amperes in intermittent duty — minutes of operation followed by rest. That combination allows smaller cores and conductors, but it demands internal insulation that is essentially free of partial discharge. A test transformer that rains picocoulombs contaminates the PD measurement of the object under test. So the high voltage winding is sectioned into carefully graded layers, with electrostatic shields and corona electrodes at the ends, and the unit is oil-immersed, cast in epoxy, or built into a self-insulating fibreglass tank.

Short-circuit impedance is deliberately higher than in a power transformer. It limits the current at the instant the object breaks down, which is a normal and expected event in a breakdown test, and it protects the winding. The price is a voltage change under capacitive load, and note the sign: it is negative. The leading capacitive current through the series reactance produces a drop of opposite sign, so the terminal voltage rises above the no-load value. On large loads that swell has to be compensated, by primary taps, by shunt compensating reactors, or simply by calibrating the output voltage against command voltage curve. Ignoring the effect is one of the most common causes of accidental overvoltage in inexperienced laboratories.

The capacitive load, not the voltage, sets the rating

The golden rule of AC sizing is the apparent power demanded by the capacitive load: S = ω·C·U² = 2π·f·C·U². In specification work the practical form is P ≈ k·ω·C·U², where k is a design safety factor typically between 1.1 and 1.5 covering capacitance tolerance, harmonics and regulation reserve. The corresponding test current is I = ω·C·U.

Three worked cases at 60 Hz calibrate the intuition. A distribution glass insulator at roughly 30 pF tested at 100 kV needs about 113 VA — any bench source handles it. A 245 kV condenser bushing at roughly 500 pF tested at 200 kV asks for about 7.5 kVA, still comfortable. But one kilometre of 138 kV extruded cable at roughly 0.25 µF tested at 128 kV demands about 1.54 MVA, with a test current around 12 A. That is out of reach for a conventional test transformer and is the natural territory of the resonant circuit. Two refinements complete the job. First, the voltage rise across the source impedance means the measuring system belongs on the high voltage side and never inferred from the primary. Second, thermal duty: a production line running dozens of applications per hour needs different thermal sizing from a one-off type test.

Cascades buy voltage and pay in utilisation

Building a single transformer for 600 kV is possible, but insulating it for full voltage makes the unit disproportionately expensive, heavy and hard to transport. The classic answer, in service since the 1920s, is the cascade: two or three identical units, each insulated only for its stage voltage V, stacked electrically. Each stage secondary adds to the tank potential of the next stage, and the upper stage is fed by a coupling winding in the stage below. The second tank floats at V, the third at 2V, and the output reaches 3V — three 200 kV stages giving 600 kV.

The elegance carries a cost the designer has to see. If the output delivers 3P, the base transformer processes its own contribution plus all the power of the stages above it: its primary and coupling winding carry 3P. The middle carries 2P and the top carries P. Installed power is therefore 6P for 3P useful, a utilisation of 50%, and in general 2/(n+1) for n stages: 67% with two stages, 50% with three. On top of that, the equivalent series impedance grows roughly with the square of the number of upstream stages, degrading regulation and amplifying the capacitive rise. That is why cascades rarely exceed three stages, and why voltage control always acts on the base supply, never by switching between stages.

Series resonance: the physics pays the bill

If the load is a capacitor, put it in series with a reactor and let resonance do the work. At f₀ = 1/(2π√(LC)) the reactances of L and C cancel, the circuit seen by the source becomes almost purely resistive, and the voltage across the object reaches Q times the excitation voltage, where Q = ωL/R is typically between 30 and 100 in real systems. The consequence is dramatic: the active power the grid has to supply is only S/Q, that is 1% to 3% of the test apparent power. The 1.54 MVA cable test above draws something like 20 to 50 kW from the mains, which is an ordinary industrial outlet.

There are two ways to stay tuned when C changes from object to object. Tuned systems fix the frequency at mains value and adjust an air gap in the reactor core with a servomotor: robust, but with moving parts and a heavier reactor. Frequency-tuned systems keep the reactor fixed and sweep a static converter, typically 20 to 300 Hz, until f₀ is found. Tuning is automatic, the reactor is substantially lighter per kVA, and there are no moving parts. IEC 60840 for 30 to 150 kV extruded cables and IEC 62067 for 150 to 500 kV explicitly recognise on-site testing in that 20 to 300 Hz band, which made the frequency-tuned system the world standard for cable commissioning. Short cables with small C resonate at the top of the band; kilometres of cable resonate near 20 Hz.

The decisive argument is what happens when the object fails. In a transformer-fed test, breakdown short-circuits the secondary and the source feeds the arc with all the current its impedance allows, until protection operates — enough energy to enlarge the fault and mask the diagnosis. In a resonant circuit, breakdown destroys the resonance condition itself: the circuit capacitance changes instantly, tuning is lost, the Q gain disappears and the voltage collapses by physics rather than by protection. Fault current stays limited to the rated circuit current and the arc energy is orders of magnitude lower. Add the waveshape benefit: the LC circuit is a filter that delivers a nearly pure sinusoid even from a distorted excitation, because harmonics simply find no gain away from f₀.

Control, waveshape and what the standard actually demands

None of the topologies is worth anything without fine voltage control. The classic element is the variable autotransformer with a sliding brush, motorised in laboratory sources to give smooth ramps from zero to full voltage. At higher powers the column regulator takes over, and modern sources use PWM static converters with an output filter, which add a degree of freedom the autotransformer never had: frequency becomes programmable, enabling frequency ramps for induced-voltage testing and automatic tuning of resonant circuits. One operating rule is not negotiable and carries over from the safety module: energisation always starts from zero, and the interlock should prevent closing the power circuit with the regulator away from its initial position.

IEC 60060-1 is the standard that validates the test, and edition 4.0 was published in 2025, cancelling and replacing the 2010 third edition. Its central requirements for AC test voltage are these. The waveshape must be sinusoidal with a peak to RMS ratio equal to √2 within ±5%, and it is that ratio, not total harmonic distortion directly, that the laboratory verifies. The test value is defined as the peak divided by √2, so peak measurement is the reference. Frequency for a standard power-frequency test lies between 45 and 65 Hz, with extended bands such as 20 to 300 Hz where the product standard allows them. The level must be held within ±1% for tests up to 60 s, and ±3% for longer durations, which requires active regulation and high-side measurement with uncertainty no worse than 3% under IEC 60060-2. Above 75% of the test value the voltage rises at about 2% of U per second — fast enough not to stress the object needlessly, slow enough to read and react.

Syllabus

• Test transformer versus power transformer: ratio, current, duty cycle, internal partial discharge requirement

• Insulation construction: graded layers, electrostatic shields, corona rings, oil-filled, cast-resin and self-insulating tank designs

• Short-circuit impedance as a deliberate current limiter at object breakdown

• Capacitive voltage rise under load and its compensation by taps or shunt reactors

• Sizing from the load: S = ωCU², I = ωCU, and the design safety factor k

• Worked cases: a distribution insulator at 100 kV, a 245 kV condenser bushing, one kilometre of 138 kV extruded cable

• Cascade transformers: coupling winding, floating tanks, and the 3V output from three V stages

• Power distribution in a cascade: 6P installed for 3P useful, and impedance growing roughly with n²

• Series resonance: f₀ = 1/(2π√(LC)), quality factor Q, and grid power of S/Q

• Tuned-reactor and frequency-tuned resonant systems; the 20 to 300 Hz band accepted by IEC 60840 and IEC 62067

• Self-protection at breakdown: detuning collapses the voltage by physics, not by protection

• Voltage control from variable autotransformer to static converter with programmable frequency

• IEC 60060-1 waveshape, frequency, tolerance and ramp requirements

• The eight-step procedure for specifying an AC test circuit

Laboratory work

Build the calibration curve of a test transformer and run a withstand test on insulators, following the eight-step specification routine. First, the pre-energisation inspection and interlock check from the safety module. Second, measure the insulator capacitance on a bridge and record C in pF. Third, compute the expected I and S before energising. Fourth, take ten points of command voltage against voltage measured on the high-side divider, once with no load and once with the insulator connected, so the effect of the capacitive load on the effective ratio appears as a number rather than an assertion. Fifth, check the waveshape on the oscilloscope: compute peak/RMS and compare against √2 ±5%. Sixth, run the one-minute withstand test at the level the insulator standard requires, ramping at 2% per second above 75% of U. Seventh, issue the verdict with the curve attached and every deviation commented. Directed extension: for a 500 kVA conventional source, determine the largest cable length at 150 kV and 0.5 µF/km you could test, then specify the resonant alternative and the grid power it would draw at Q = 60.

Key takeaway

During a test the measured object voltage exceeds what the source should deliver. That is not automatically a measurement error. It can be the capacitive rise across the source series reactance, a few per cent but enough to invalidate a test read from the primary. Or it can be accidental series resonance between the object capacitance and the source leakage inductance, in which case the voltage runs away with gain Q. Measure on the high voltage side, always, and calculate f₀ for source plus object before you energise. What is an accident in a conventional hipot is, domesticated, the operating principle of the resonant one.

Module 3 of the Atlas Energy Academy · Generating & Measuring · about 6 hours · Video series plus hands-on AC bench practice (part A) · Prerequisites: Module 2 — Lab Safety and Anatomy