Parasitic Phenomena and Test Installations

The whole laboratory is one circuit, including the parts nobody drew.

Every test engineer has heard the question from a customer: why was the same transformer accepted in laboratory A and rejected in laboratory B? Set aside the legitimate difference in measurement uncertainty, which Module 10 handles, and the answer is almost always parasitic phenomena. These are the electrical, thermal and electromagnetic effects that never appear on the single-line diagram of the test but take part in it as concretely as the generator and the object. Modules 3 through 8 treated the source, the divider and the instrument as ideal blocks joined by impedance-free lines. This module removes that fiction.

Between the source and the recorder there are electrodes with finite radii of curvature that go into corona, connections whose contact resistance grows quietly until it becomes a hot spot, cable loops that behave as inductors and antennas, grounded walls that distort the electrostatic field, sources that deliver deformed sine waves, and earth meshes carrying stray current. The thesis is easy to state and laborious to practice: the whole laboratory is a single circuit. Stray capacitance between the HV bus and a grounded wall is a component of that circuit even though nobody drew it. The inductance of the loop formed by the return conductor is a component. The contact resistance of a poorly torqued terminal is a component. The competent test engineer sees those invisible components, estimates their order of magnitude, and decides whether they move the result beyond the tolerance the standard allows.

What you will be able to do

• Identify and mitigate corona, hot spots and oscillation in test setups, applying the Peek criterion to predict corona onset gradient as a function of curvature radius and specifying appropriate grading electrodes.

• Explain electric field distortion and harmonic distortion and quantify their effect on the test result, using the Schwaiger utilization factor, minimum clearances to walls and ceiling, and the AC waveform requirement of IEC 60060-1:2025 that peak divided by √2 agree with the RMS value within ±5%.

• Design grounding and measurement cable routing that minimizes interference: single-point reference, minimum loop area, continuous shielding, impedance matching and correct coaxial termination.

• Audit a complete setup before the test: cable continuity and integrity, scale and attenuator settings, ground references, background noise in pC per IEC 60270, and conformity of the circuit to the procedure.

• Diagnose the classic setup errors from photographs and oscillograms: loose cables, floating references, attenuator saturation, wrong scale factors, ground loops and missing corona rings.

Corona and the Peek criterion

Corona is a self-sustaining partial discharge in gas next to an electrode of small curvature radius, where the local field exceeds the ionization field of air but the gap is too long for a full breakdown channel. The avalanche converts to a streamer when space charge at its head reaches the Raether-Meek criterion, around 10⁸ electrons or αx of 18 to 20, and because the field falls off quickly around a curved electrode the streamer dies within a few centimeters. Polarity sets the morphology: positive half cycles give longer filaments with pulses of hundreds of milliamperes and tens to hundreds of nanocoulombs, negative half cycles give Trichel pulses that are short, regular and repeat at tens of kilohertz to megahertz. That regularity makes them very effective at polluting the radio spectrum.

Peek's empirical law is still the first design tool for any HV setup. For coaxial cylindrical geometry with conductor radius r, the critical surface field in RMS terms is E_c = 21.1·δ·m·(1 + 0.301/√(δ·r)) in kV/cm with r in cm. The 21.1 kV/cm RMS, about 29.8 kV/cm peak, is the reference gradient for air at sea level; δ is relative air density, so a laboratory at 900 m sits near δ = 0.90 and loses the same proportion; m is the surface factor, 1 for polished and 0.6 to 0.8 for rough, oxidized or wet surfaces, since a single droplet can cut the local critical field by 40%. The 0.301/√(δr) term looks paradoxical, but since the field at a curved electrode grows as U/(r·ln(D/r)), inception voltage still collapses with radius: a 0.5 mm point coronas at a fraction of the voltage a 250 mm toroid withstands.

Corona losses, RIV, and the partial discharge noise floor

Above the critical voltage, dissipated power grows with the square of the overvoltage. Peek's loss formula for a cylindrical conductor at 50/60 Hz is P = (241/δ)·10⁻⁵·(f + 25)·√(r/D)·(U − U_c)² in kW/km per phase with U in kV. That (U − U_c)² dependence has two consequences in the laboratory. Energetically, corona on a badly built bus is real additional load on the source during long tests, distorting the test current. Metrologically, corona is a pulsed non-sinusoidal current drawn from the most sensitive node in the circuit, the HV node, so it superimposes on any current, dissipation factor or partial discharge measurement made there. A tan delta test with corona in the connection reports fictitious dielectric loss in the object. A PD test with external corona is blind to the internal discharges that actually matter.

Each corona pulse has a nanosecond rise time, making it a wideband generator radiating from hundreds of kilohertz to tens of megahertz. Measured deliberately that is radio influence voltage, read in µV across 300 Ω with a quasi-peak detector at 1 MHz under CISPR TR 18 and IEC 60437:2023, with receiver requirements from CISPR 16-1-1; there is no universal RIV limit, the product standard or contract sets it. Inside the laboratory the same phenomenon has a different victim: PD background noise. An IEC 60270 instrument integrates current pulses typically from 100 to 400 kHz and cannot by itself tell an internal cavity pulse from a corona pulse three meters away. Hence the rule: before any PD measurement the whole setup must be corona-free above the maximum test voltage, verified by energizing without the object and confirming background below 50% of the specified magnitude, with good practice at 20 to 30%.

Hot spots: Holm's constriction resistance and what thermography sees

Two apparently flat metal surfaces actually touch at a small number of microscopic asperities, the a-spots of Holm's theory. All the circuit current is forced to squeeze through them, and the constriction resistance of a circular contact of radius a is R_c = ρ/(2a). For copper with an aggregate a-spot of 0.1 mm, R_c is about 85 µΩ — invisible to an ordinary multimeter, decisive when hundreds or thousands of amperes flow. On top of the constriction sits film resistance from oxides, sulfides and contaminants, orders of magnitude more resistive than the metal. Degradation feeds itself: heating accelerates oxidation, which raises resistance, which raises P = R_c·I², which drives differential expansion and creep that relax the clamping pressure, which reduces the number of a-spots. That is the cycle that turns a 50 µΩ joint into a 500 µΩ joint running 80 K above ambient within months. Holm's voltage-temperature relation, θ_max² = θ_amb² + U_c²/(4L) with the Lorenz number L = 2.45·10⁻⁸ V²/K², shows why: a drop of only 0.3 to 0.4 V across the contact softens copper near 190 °C and about 0.43 V melts it locally. Power connections are therefore designed for millivolt drops.

Thermography is the audit instrument, with three cautions. Emissivity: polished copper sits at ε of 0.03 to 0.1, a thermal mirror that fools the camera, so apply matte paint or known-emissivity tape at the point of interest. Load: inspect at representative current, at least around 40% of rating. And compensate ambient and reflected temperature. ANSI/NETA MTS-2023 Table 100.18 gives the severity classes: 1 to 3 K between similar components under similar load is a possible deficiency to monitor, 4 to 15 K is a probable deficiency to repair as schedule permits, and above 15 K is a major deficiency requiring immediate repair. In dielectric testing the effect is subtler: an HV connection with abnormal resistance is almost always mechanically loose, and loose means micro-gaps that generate contact discharges, appearing as erratic pulses sensitive to vibration.

Stray oscillation and the normative treatment of overshoot

Every length of conductor has inductance — roughly 1 µH per meter for an insulated cable away from the ground plane, 0.5 to 1.5 µH/m depending on geometry. Every conducting surface facing another has capacitance: a 2 m divider sees the ceiling and walls through some tens of picofarads, and the bus joining generator to object adds more. With the intentional elements of the circuit these form LC meshes with natural frequencies typically between hundreds of kilohertz and tens of megahertz, since 5 µH with 100 pF resonates at 7.1 MHz. Every fast transient excites them — a gap firing, a voltage collapse at breakdown, a source switching event — and the oscillatory response superimposes on the signal you want. The classic sources are the trigger loop between generator stages, the measuring loop formed by HV bus plus divider plus ground return (several square meters is common in careless setups, and loop inductance is proportional to area), the object capacitance resonating with the series inductance of the front circuit, and reflections in badly terminated measurement cables.

IEC 60060-1 accepts that some oscillation is unavoidable and disciplines its evaluation. The record splits into a base curve, the mean double exponential, and a residual carrying oscillation and overshoot. Relative overshoot β is the difference between recorded peak and base curve peak referred to the latter, limited to 10%. Because insulation does not respond equally to all frequencies, the standard filters the residual with k(f) = 1/(1 + 2.2·f²), f in MHz, and adds it back. The practical consequence answers this module's title question: two setups with the same generator report different test voltages if one rings at 300 kHz, where k ≈ 0.83, and the other at 3 MHz, where k ≈ 0.048. Mitigation is damping resistors at generator output and divider input, minimum loop area, short direct connections to the divider, and a low-voltage step response check of the complete assembly — persistent ringing there is a fault of the installation, not the divider.

The wall takes part in the test, and so does the source waveform

The field around an HV setup is not the field of the catalogue geometry. Walls, ceiling, floor, metal structures and anything else conductive and grounded deform the equipotential surfaces. There are two consequences. Disruptively, moving the object toward a wall compresses the equipotentials in the object-to-wall gap and lowers the external flashover voltage, so a withstand test can fail on an air flashover that would never happen in the standard arrangement. Metrologically, stray capacitance from divider to wall changes the voltage distribution along the divider column and therefore its ratio at high frequency. The quantity that measures uniformity is the Schwaiger utilization factor η = E_mean/E_max = (U/d)/E_max, close to 1 in near-uniform fields and small in point-plane geometry. Bringing grounded mass closer reduces η. The practical rules are a clearance to walls and ceiling of at least 1.5 times the shortest disruptive distance in the test, and, given that large point-plane gaps withstand roughly 500 kV peak per meter for positive lightning impulse and 380 to 400 kV/m for AC peak, a rule of thumb of 2 to 3 m clear per megavolt of test voltage in every direction, ceiling included.

The value that defines an AC withstand test is the peak divided by √2, because breakdown responds to the peak, while ordinary instruments read RMS. IEC 60060-1 therefore requires peak over √2 to match the RMS value within ±5%, at 45 to 65 Hz. Note the precision: the 2025 edition sets no general THD limit, the ±5% peak relation being the dominant criterion, with harmonic limits left to product committees; in practice keeping THD near or below 5% meets it with margin. Harmonics come from test transformer saturation flattening the crest, from thyristor or variac regulators, and from the supply. There is also a reverse effect: a large capacitive object on an inductive source selectively amplifies harmonics near the series resonance of source inductance with object capacitance, inflating the real peak several percent above what an RMS voltmeter suggests.

Ground loops, shielding and the measurement cable

When the object, the attenuator and the oscilloscope are each grounded at the most convenient point of the mesh, you have created a closed circuit. It contaminates the signal two ways. Inductively, the loop is a turn, and the varying flux through it — the brutal dΦ/dt of a generator firing, the return current of a breakdown circulating in the mesh — induces a voltage in series with the measurement signal, tens or hundreds of volts in a loop of a few square meters near an impulse circuit. Conductively, the earth mesh is not equipotential during transients, so potential differences between distant points inject current into the cable shield, and the shield's transfer impedance converts that current into signal voltage. The classic results are phantom oscillation, baseline shift, and, on a PD meter, pulse trains synchronous with the mains that are indistinguishable from real discharges. The fix is a single-point reference: pick one reference ground, typically the divider foot or the object earth terminal, bring every measurement return to it in a star, keep the coaxial shield continuous from source to instrument, and use short, wide, low-inductance ground conductors, because inductance and not resistance dominates transient behaviour.

Estimate the couplings before you suffer them. Capacitively, i = C_p·dU/dt: 10 pF, a measurement cable 30 cm from a bus, with dU/dt of 1 kV/ns at breakdown, is 10 A injected into an unshielded cable. Inductively, u = M·di/dt, so a 10 kA/µs return with M = 0.1 µH induces 1 kV. Shield and separate against capacitive coupling; reduce loop area and cross at 90° against inductive coupling. Never run a measurement cable parallel to a power bus closer than a meter. The room shield is the last resort against externally radiated noise, reaching 60 to 100 dB in a well-built cage, but it is a placebo against conducted noise entering through the supply, against noise generated inside the room, and against internal ground loops, which are topology. Finally, the measurement cable is a transmission line: terminate in Z0, never splice with improvised connectors, and remember that swapping 10 m for 25 m without requalifying has changed the measuring system, which IEC 60060-2:2025 treats as a whole.

Syllabus

• The five families of parasitic phenomena: corona, hot spots, stray oscillation, distortion, interference

• Corona physics: avalanche, Raether-Meek streamer criterion, polarity morphology and Trichel pulses

• The Peek criterion: E_c = 21.1·δ·m·(1 + 0.301/√(δ·r)) and the effect of altitude and surface finish

• Peek's quadratic loss formula and the energetic and metrological consequences of corona

• RIV measurement per CISPR TR 18 and IEC 60437:2023, and the conflict with PD measurement

• Background noise discipline before any IEC 60270 measurement

• Anti-corona engineering: minimum radii, toroids and grading rings, surface finish, tubular bus

• Contact resistance: Holm a-spots, R_c = ρ/(2a), the film resistance and the self-feeding degradation cycle

• The Holm voltage-temperature relation and why power connections are designed for millivolt drops

• Thermography: emissivity, minimum load, and the ANSI/NETA MTS severity classification

• Stray inductance and capacitance: 1 µH/m rule, loop area, and MHz resonance

• Base curve, overshoot β and the k(f) test function as the normative treatment of oscillation

• Damping, loop minimization and step response verification of the complete installation

• Field distortion: the Schwaiger utilization factor and clearance rules of thumb

• Harmonic distortion: peak versus RMS, source saturation, and series resonance amplification

• Ground loops: inductive and conductive coupling mechanisms; single-point star reference

• Capacitive and inductive coupling estimates before the fact

• The Faraday cage: when it is mandatory and when it is a placebo

• Measurement cables as transmission lines: Z0, termination, splices, attenuation, requalification

Laboratory work

An investigative exercise: five real scenarios, each presented as photographs of the setup plus oscillograms or PD records, in which you name the setup defect, justify the observed symptom physically, and prescribe the correction. Scenario 1, the impulse that rings: a lightning impulse record with roughly 2 MHz oscillation and 18% apparent overshoot on the crest, with the divider 6 m from the object, joined by a thin catenary cable, and a 10 m ground return along the wall. Estimate the loop inductance at 15 to 20 µH, predict the oscillation frequency, evaluate β with the k(f) weighting, and reconfigure. Scenario 2, the PD that comes and goes: a cloud of pulses on the positive crests at about 200 pC in a voltage transformer test, changing when the operator touches the bus with the operating rod, with photographs showing a protruding bolt and a loose tape end at the HV terminal. Scenario 3, laboratory B fails the object: tan delta of 0.35% in one lab and 0.52% in another, where the second has the object 40 cm from the wall and the measurement cable running 8 m parallel to the return bus. Scenario 4, the clipped peak: successive impulses with the peak pinned at exactly the same reading and front times growing with level. Scenario 5, the nine o'clock noise: an 8 pC PD background that appears at 09:00 and disappears at 17:00 every working day. Deliverable: for each scenario a half-page report stating the defect, the physics of the symptom, the correction and how to validate the correction. Passing criterion is four correct diagnoses out of five with quantitative justification in at least two.

Key takeaway

The investigative format is not a teaching device, it is the working method for any anomalous result. Do not discard the record — the bad oscillogram is the richest data of the day. Classify the symptom: oscillation, drift, saturation, synchronous noise, or a value displaced by a round factor. Form a physical hypothesis and estimate its order of magnitude, because a hypothesis without a number is a guess. Test cheap first: move the cable, lift one ground, reduce the gain, darken the room. Correct one variable at a time and repeat the test. Then document it, because the laboratory's catalogue of errors is metrological capital. And keep the underlying thesis in view: the whole laboratory is a single circuit, and the test engineer is the person who can see the components that are not on the diagram.

Module 9 of the Atlas Energy Academy · The Lab as a System · about 4 hours · Video series plus investigative case studies · Prerequisites: Module 8 — High Current Generation and Measurement