Very Fast Transient and Enclosure Voltage Study for a Gas-Insulated Substation

1. Executive Summary

Keentel Engineering Solutions was retained by the Owner of a 345 kV gas-insulated substation (GIS) to determine why the station's secondary systems misbehaved whenever a disconnector was operated, and whether the phenomenon responsible also threatened the primary insulation. The presenting symptoms were three. Numerical relays and station computers recorded spurious status changes, communications dropouts and, on nine occasions in the fourteen months preceding Month 0, protection misoperations requiring reporting under NERC PRC-004. Field personnel reported perceptible and in two cases startling electric shocks on contact with enclosure surfaces during disconnector operation, in a yard where nothing was energized that anyone could touch. And the Owner's asset group had begun to question whether the two 345/138 kV transformer banks connected directly to the GIS through SF6-to-oil bushings were accumulating insulation damage that no routine test would find. The three symptoms have one cause. A gas-insulated substation is an electrically compact coaxial system in which a traveling wave crosses a bay in tens of nanoseconds, the bus surge impedance is 50 to 80 Ω rather than the 300 to 400 Ω of an overhead line, and a disconnector's slow-moving contacts strike and restrike hundreds of times during a single operation. The result is a class of transient — very fast transient overvoltage (VFTO) inside the enclosure, and transient enclosure voltage (TEV) on the outside of it — with rise times of a few nanoseconds and oscillation frequencies in the tens of megahertz. These phenomena do not exist in an air-insulated substation in any comparable form. They are invisible to load flow, to short-circuit analysis, to conventional switching studies, and even to a standard lightning study, because every one of those methods uses a time step and a model bandwidth that filters them out before they can appear. The study was executed in PSCAD/EMTDC with an independent EMTP cross-check, using a distributed-parameter representation of 214 individual GIS sections, a nanosecond-scale spark model for the disconnector arc, a measured high-frequency capacitive model of the transformers, and a separate transmission-line model of the enclosure with respect to the station ground grid. The time step was 0.5 ns over a 20 µs window, and 9,000 Monte Carlo shots were used to sample restrike instants and trapped-charge states. The results were unambiguous. VFTO reached a statistical 2 percent value of 2.34 pu of peak line-to-ground voltage at the transformer terminal and 2.11 pu on the bus, ringing at 12.8 MHz with a 4.6 ns rise time — roughly 260 times faster than the 1.2/50 µs lightning impulse against which the equipment's 1050 kV BIL was demonstrated. TEV measured 8.4 kV peak at a bushing flange with a dominant 6.1 MHz component. Induced voltage on unshielded secondary cabling reached 3.1 kV, exceeding the 2.5 kV oscillatory surge withstand crest of IEEE Std C37.90.1 at a frequency an order of magnitude above what that test applies. The recommended package — 500 Ω damping resistors on the four disconnectors adjacent to the transformer bays, a rebuilt enclosure bonding and equipotential mesh scheme, shielded and double-bonded secondary cabling with parallel ground continuity conductors, and an interim operational restriction — reduced VFTO to 1.62 pu, TEV to 1.9 kV, and secondary induced voltage to below 300 V. Misoperations ceased. The shock reports ceased. The Owner retained the validated model and added VFTO and TEV requirements to its GIS procurement standard. Figures presented are representative of the delivered study and have been generalized to protect client confidentiality. Study at a Glance

2. Background and Study Drivers

The station is a nine-bay 345 kV indoor GIS in a double-bus, single-breaker arrangement, serving four transmission lines through SF6-to-air bushings, two 345/138 kV transformer banks through SF6-to-oil bushings, one bus coupler bay and two bays reserved for future extension. The GIS replaced an air-insulated yard whose footprint the Owner could not extend, and the compactness that justified the replacement is precisely the property that produced the problem. The commercial and compliance stake had three parts. The first was regulatory. Protection misoperations are reportable and correctable events under NERC PRC-004, and a station generating them at a rate approaching one every six weeks, with no fault present and no identified relay defect, is a standing compliance exposure. The Owner's protection group had already replaced two relays, reflashed firmware on four more, and rerouted one cable, none of which changed the rate. The second was personnel safety. Three shock reports had been logged by switching crews, two describing an involuntary startle reaction on contact with a bushing flange handrail. No injury resulted. The Owner correctly treated this as a serious finding rather than a nuisance, because the hazard in a TEV event is not electrocution — the discharge is energy-limited and brief — but the startle-and-fall response of a person on a ladder, a platform or a walkway, which is the mechanism OSHA 1910.269 and NFPA 70E exist to prevent. Operations had responded by restricting disconnector operation to periods when the affected bay was unoccupied, an unsustainable constraint on switching. The third was asset risk. VFTO stress on a transformer is not detectable by any conventional acceptance test. It does not show on a power-factor test, it does not necessarily show on a routine dissolved gas analysis until damage is well advanced, and it is not covered by the BIL demonstration because the BIL test waveform is far slower than the stress. The Owner had two banks with an eighteen-month replacement lead time and a spare policy covering one phase. A quantified answer on winding stress was worth considerably more than the study cost. A fourth driver emerged during scoping. The Owner intended to extend the GIS into the two reserved bays and had a procurement specification that said nothing about VFTO, nothing about TEV, and nothing about damping resistors. Whatever the study found, the specification needed to change before the extension was tendered.

3. Study Objectives and Scope of Work

1. Characterize the disconnector restrike process at this installation, including the number of restrikes per operation, the distribution of trapped-charge states, and the resulting statistical distribution of very fast transient overvoltage.
2. Determine VFTO magnitude, dominant frequency and rise time at every dielectrically significant location in the GIS, and compare against the equipment insulation levels with explicit treatment of the fact that a nanosecond front is not covered by the BIL demonstration.
3. Determine the voltage distribution imposed on the transformer high-voltage winding by the VFTO front, quantify the departure from uniform distribution across the line-end turns, and identify part-winding resonance exposure.
4. Determine transient enclosure voltage at every enclosure discontinuity — bushings, viewing ports, insulating flanges, cable terminations, expansion joints — and assess personnel shock exposure and interference potential.
5. Quantify radiated and conducted coupling from the GIS into secondary cabling, and compare induced voltages against the surge withstand capability levels of IEEE Std C37.90.1 and the immunity levels of the IEC 61000-4 series.
6. Verify circuit breaker transient recovery voltage against IEEE Std C37.06, including terminal fault, short-line fault and out-of-phase duties, and evaluate cable and transformer energization and lightning ingress through the overhead-to-GIS interface.
7. Validate the model against a field measurement campaign using capacitive dividers and high-bandwidth acquisition, to a stated acceptance criterion.
8. Compare mitigation options on technical merit, order-of-magnitude cost, schedule and residual risk; recommend a package; and supply specification language for the Owner's GIS procurement standard.

Scope and Deliverables

Two exclusions require comment. The study did not build a detailed lumped-element internal winding model of the transformers, because the manufacturer's winding geometry was not available and a model built on assumed geometry would have produced a precise-looking number with no evidential value. Instead the study quantified the initial capacitive distribution from measured terminal capacitances, which is defensible, and stated the resonance screening as a screening result rather than a stress calculation. The study also did not redesign the station ground grid for power-frequency fault performance; the grid was adequate at 60 Hz and the finding concerned its behavior at megahertz frequencies, which is a bonding problem rather than a grid problem. The Owner accepted both exclusions in writing.

4. Regulatory and Standards Basis

The regulatory drivers are protection performance and facility rating integrity. NERC PRC-004 requires identification and correction of protection system misoperations; a station producing misoperations from an electromagnetic interference mechanism has an uncorrected cause, and closing those events required identifying it. NERC PRC-005 governs maintenance and testing of protection systems, and the Owner's testing program could not have detected an interference-induced misoperation because the relays were functionally sound. NERC FAC-008 requires facility ratings consistent with equipment ratings; equipment subjected repeatedly to a dielectric stress outside the envelope in which its rating was demonstrated is not clearly within rating, and the study was scoped in part to establish whether that was the case here. Standards and Regulatory Register

US practice governs throughout. Insulation levels are stated as BIL from the preferred-level tables of IEEE Std C57.12.00 and IEEE Std C37.06, not against an IEC highest-voltage-for-equipment ladder. Immunity is assessed primarily against IEEE Std C37.90.1, with the IEC 61000-4 series cited where it covers a stress the IEEE test does not, which in this study it does. Voltages are referred to a base of 295.6 kV, the peak line-to-ground value at the 362 kV maximum system voltage of ANSI C84.1.

5. System Modeling and Data Development

5.1 Data sources and gap closure

A VFTO study is unusually data-hungry in one specific respect: it requires physical dimensions. Every straight run, every elbow, every disconnector chamber, every spacer, every bushing and every enclosure section must be represented by its own surge impedance and its own travel time, and both derive from geometry. A single-line diagram is close to useless. The study therefore began with a dimensional survey. Model Data Sources and Gap Closure

The bonding survey and the cable trace were the two most valuable data activities, and both found conditions the drawings did not show. Fourteen of the 96 enclosure bonding points used straps between 9 and 13.5 feet long where the design intent had been under 3 feet. Nine of the 41 traced secondary circuits ran within 12 inches of a bushing flange for more than 16 feet, and six of those had cable shields bonded at one end only, a practice that is correct for low-frequency electrostatic screening and actively harmful at megahertz frequencies.

5.2 Model construction

The GIS was represented as a network of lossless distributed-parameter transmission lines. This is the correct formulation and effectively the only correct one: a coaxial gas-insulated bus is an almost ideal TEM structure with negligible dispersion over the band of interest, so its behavior is fully described by a surge impedance and a travel time, and both follow from the conductor and enclosure diameters. Surge impedance was computed as Z = 60·ln(D/d) for each section, where D is the enclosure inner diameter and d the conductor outer diameter, giving values from 48 Ω in the disconnector chambers, where the local conductor is enlarged, to 78 Ω in the spacer regions. The main bus runs came out at 62 Ω. Propagation velocity in SF6 at the operating density is within 2 percent of the speed of light, and 0.98c was used, giving a travel time of 3.40 ns per meter, or 1.04 ns per foot. Representative GIS Section Parameters

The contrast in the last two rows is the entire physics of the problem in one table. A wave arriving at the transition from a 62 Ω bus into a 340 Ω overhead line sees a reflection coefficient of +0.69 and nearly doubles; arriving at a transformer bushing terminated in a few nanofarads, it sees an effective open circuit at these frequencies and doubles outright. In an air-insulated substation the same wave travels between elements separated by hundreds of feet at 984 ft/µs and its reflections are smeared across a front measured in microseconds. Here they are separated by nanoseconds and arrive coherently. The disconnector was represented by a spark model rather than an ideal switch. When the contacts approach and the gap breaks down, the arc resistance does not step to zero; it collapses from an essentially infinite pre-breakdown value to a residual arc resistance with a time constant of the order of one nanosecond. The model used R(t) = R₀·e^(−t/τ) + R_arc with τ = 1.1 ns, R_arc = 0.7 Ω and R₀ = 10¹² Ω. This matters more than it appears to. An ideal switch produces a mathematically infinite rate of rise, which the solver then limits to one time step, making the computed rise time an artifact of the time step rather than a property of the system. With the spark model the rise time is a physical result, and it is what the study reports. Breakdown initiation used a gap-voltage criterion with the withstand of the opening gap rising in proportion to contact separation. The measured contact speed was 11 inches per second over a 10-inch stroke, an operating time of about 0.89 s, and the SF6 gap withstand gradient at the operating density gives a strength recovery rate of roughly 1.2 kV per microsecond of travel. Against a 60 Hz source traversing its full range every 8.33 ms, that recovery rate is glacial, and the consequence is the restrike train described in Section 7.2. The transformers were represented by a measured high-frequency capacitive network rather than by leakage impedance. At 12.8 MHz a transformer is not an inductance; it is a capacitance with a complicated internal structure. The measured values were a terminal-to-ground capacitance of 3,850 pF and an equivalent series winding capacitance of 210 pF, with the bushing contributing 480 pF in parallel. The model reproduced measured terminal impedance within 8 percent from 10 kHz to 5 MHz and within 15 percent to 20 MHz. The enclosure was modeled as a separate transmission-line network referenced to the station ground grid, with a computed enclosure-to-ground surge impedance of 130 Ω for the elevated runs. Every discontinuity — bushing flange, viewing port, insulating flange at the cable terminations, expansion joint — was represented as a coupling point between the internal coaxial system and the external enclosure system, with the coupling coefficient derived from the geometry of the aperture. Bonding straps were modeled as inductances at 0.32 µH per foot from the surveyed lengths, and the station grid as a frequency-dependent impedance from the injection test. Time step was 0.5 ns. This was verified by rerunning the governing cases at 0.2 ns, which changed peak VFTO by 0.9 percent and dominant frequency by 0.4 percent. The simulation window was 20 µs for VFTO and TEV cases, long enough to contain the full ringing decay and any reflection returning from the remote end of the shortest overhead line, and 100 ms for the TRV and energization cases at a 2 µs step.

5.3 Why conventional models fail here

It is worth stating explicitly what a conventional model does to these phenomena, because the Owner's first reaction was that the GIS had been studied at procurement and nothing had been found. A 60 Hz load flow or short-circuit model represents the entire GIS as a node. Its physical extent is zero and its travel times do not exist. A conventional switching study at a 10 to 50 µs time step has a Nyquist limit of 10 to 50 kHz; a 12.8 MHz oscillation is not attenuated by such a model, it is simply absent, and the solver will return a smooth, plausible, entirely wrong answer. Even a competently built lightning study, at a 2 ns step with frequency-dependent line models, will usually miss VFTO — not because of bandwidth but because of topology. A lightning model represents the incoming line in detail and the station in outline, often lumping the GIS into a few sections with a single capacitance. It has no disconnector spark model, no enclosure-to-ground network, and no reason to look at internal reflections. The procurement-stage study had been exactly this, and it had been adequate for the question it was asked.

5.4 Model validation

Model Validation and Benchmarking

The EMTP cross-check was performed because the recommendation carried a seven-figure retrofit. An independent engineer rebuilt the case from the dimensional survey without sight of the PSCAD results.

6. Study Methodology and Assumptions

6.1 Analytical approach

The study proceeded in five stages. First, a deterministic single-restrike analysis established the mechanism and identified the locations at which VFTO and TEV were highest. Second, a Monte Carlo campaign sampled the restrike instant and the trapped-charge state to produce a statistical distribution rather than a single number. Third, the transfer of the resulting front into the transformer winding was evaluated. Fourth, the enclosure and secondary-circuit consequences were computed. Fifth, mitigation options were modeled in the same framework so that their comparison used consistent physics rather than vendor claims.

6.2 Tool selection

PSCAD/EMTDC was the primary tool, chosen for its distributed-parameter line models, its handling of very small time steps over long windows, and its scripted Monte Carlo facility. EMTP provided the independent cross-check on the governing cases. CDEGS supported the frequency-dependent grid impedance derivation between measurement points. Python was used for post-processing: extraction of peaks, rise times and dominant frequencies from 9,000 result files, and the fast Fourier transforms used to compare simulated and measured spectra.

6.3 Cases and scenarios

Case and Scenario Matrix

The Monte Carlo sampling covered two random variables. The first is the instant within the 60 Hz cycle at which each restrike occurs, which sets the source-side voltage. The second is the trapped charge remaining on the disconnected section from the previous restrike, which is not independent of the first but was sampled jointly from a distribution derived from the sequential restrike simulation described in Section 7.2. The worst case, and the one the study reports, is a restrike occurring when the source side is at positive peak while the trapped charge on the isolated section is negative — the two voltages add across the gap rather than subtract.

6.4 Acceptance criteria

Acceptance Criteria

The 2.0 kV TEV criterion deserves explanation because no US standard states one. TEV shocks are capacitive discharges of very low energy and are not an electrocution hazard; the mechanism of concern is the startle reaction. Published industry experience places the perception threshold in the range of 1 to 2 kV and the reliable startle threshold above about 3 kV. The Owner and Keentel agreed 2.0 kV as a criterion on the basis that it sits at the top of the perception band and well below the startle band, and that a criterion tied to human response is more defensible than one tied to an equipment test level.

6.5 Assumptions and limitations

The model assumed lossless GIS sections. Losses in the coaxial bus are genuinely small over 20 µs, and neglecting them is conservative: the computed ringing decays more slowly than the measured, as Section 11 confirms. The spark model parameters were taken from published industry values rather than measured on this disconnector, since measuring a nanosecond resistance collapse inside an SF6 chamber was not practical; sensitivity to τ was tested and is reported in Section 8. The transformer model represents terminal behavior and the initial capacitive distribution correctly but cannot compute internal turn-to-turn stress, which the study states as a limitation rather than obscuring. Coupling coefficients at enclosure discontinuities were derived geometrically and then calibrated against the field measurement at one bushing flange, with the calibration factor applied uniformly — a simplification the study documents.

7. Analysis and Results

7.1 Why gas-insulated substations are different

Four properties, each individually modest, combine to produce a phenomenon that has no air-insulated equivalent. The first is electrical compactness. A bay of this GIS is about 39 feet of bus from breaker to bushing. At 1.04 ns per foot, that is a 41 ns one-way travel time and an 82 ns round trip. Any event whose front is faster than 82 ns will see the bay as a transmission line with reflections, not as a lumped node. In an air-insulated 345 kV yard the equivalent path is 200 to 330 feet, a round trip of 400 to 660 ns, and the events that occur there have fronts measured in microseconds. Compactness does not make GIS quieter; it makes it resonant at a higher frequency. The second is low surge impedance. At 62 Ω the GIS bus is a stiff, low-impedance system. A given injected voltage step drives a current five to six times larger than the same step on a 340 Ω overhead line, and the impedance mismatch at every interface to the outside world is severe. The third is the disconnector. A GIS disconnector is not a fast device and is not meant to be. It moves at about 11 inches per second and takes the better part of a second to complete its stroke. Its function is isolation, and its interrupting capability is limited to the capacitive current of the section it disconnects — a small current that it nonetheless cannot clear cleanly, because the gap strength recovers far more slowly than the source voltage moves. The fourth is the coaxial geometry itself, which confines the internal wave almost perfectly, so that the only paths to the outside are the discontinuities — and every wave that reaches one couples energy onto the enclosure exterior.

7.2 The restrike train

Consider a disconnector opening a short section of bus that has no other source. As the contacts separate, the section is left with whatever charge it held. The source side continues to follow the 60 Hz waveform. The voltage across the gap is therefore the difference between a moving source voltage and a static trapped voltage. When that difference exceeds the gap's instantaneous withstand, the gap breaks down, a very fast transient is launched into both sides, the section is recharged to a new voltage close to the instantaneous source voltage, and the gap recovers. The contacts have meanwhile moved a fraction of a millimeter. The cycle repeats. Each restrike leaves a new trapped charge. Early in the stroke, when the gap is short, restrikes occur near the zero crossings of the difference voltage and the trapped charge stays small. As the gap lengthens, restrikes require larger difference voltages, occur further from the crossing, and leave larger trapped charges of alternating sign. The last restrike of the sequence is the worst, because it occurs at the largest difference voltage the gap can no longer hold. Measurement over eighteen operations found between 168 and 340 restrikes per operation, with the maximum on a bus-side disconnector opening the longest isolated section. The theoretical worst case is a trapped charge of −1.0 pu meeting a source at +1.0 pu, a gap voltage of 2.0 pu. The measured and simulated distributions did not reach that bound; the governing sampled case had a trapped charge of −0.72 pu against a source at +1.00 pu, a gap step of 1.72 pu. Closing operations produce the same mechanism in reverse, with prestrikes rather than restrikes; the counts were similar and the magnitudes slightly lower, because during closing the gap is shortening and the strike occurs earlier relative to the difference voltage. Restrike Characteristics, Measured and Simulated

7.3 Very fast transient overvoltage results

When the gap breaks down, the 1.72 pu step launches two traveling waves, one into each side, with amplitudes set by the impedance division at the gap. Each wave then reflects at every discontinuity it meets. At the open end of the isolated section the reflection coefficient is +1 and the wave doubles. At the transformer bushing, terminated in a few nanofarads that look like an open circuit at 12.8 MHz, it also very nearly doubles. At the junction to a 340 Ω overhead line the coefficient is +0.69. The superposition of the trapped charge, the incident wave and these reflections is the VFTO. VFTO Results, Statistical 2 Percent Values, Pre-Mitigation

The transformer terminal is the worst location for a structural reason: it is a near-open termination at the end of a section whose length places its quarter-wave resonance at 12.8 MHz. A quarter-wave at 12.8 MHz in this medium is 19.2 ft, and the bus run from the transformer disconnector to the bushing is 19.4 ft. The GIS was, unintentionally, tuned. Against a 1050 kV BIL, 2.34 pu is 692 kV, a margin that looks comfortable. It is not the relevant comparison. The BIL demonstration uses a 1.2/50 µs impulse. This front rises in 4.6 ns, roughly 260 times faster. Insulation coordination as codified in IEEE Std 1313.1 recognizes a very-fast-front class precisely because the standard impulse tests do not cover it, and no routine factory or field test on any of this equipment applied a comparable waveform. The correct statement is not that the equipment passes with margin; it is that the equipment's withstand at this waveform is unverified, and the stress is being applied several hundred times per switching operation.

7.4 Transfer into the transformer winding

For a front this fast, the initial voltage distribution along a transformer winding is governed entirely by capacitance, not inductance. The distribution parameter is α = √(C_g/C_s), where C_g is the total winding-to-ground capacitance and C_s the equivalent series capacitance. The measured values give α = 4.3. For a step applied to the line end of a winding with a grounded neutral, the initial distribution follows a hyperbolic function of position, and with α = 4.3 the first 10 percent of the winding takes 35 percent of the applied voltage — 3.5 times what a uniform distribution would give. At the pre-mitigation VFTO of 2.34 pu, that is 242 kV across the first tenth of the winding. The turn-to-turn and disc-to-disc stress in that region is correspondingly concentrated, and it is concentrated in exactly the part of the winding where interturn insulation is thinnest relative to the stress it normally sees. Two further effects were assessed. The first is part-winding resonance. The measured terminal sweep showed winding resonances at 148 kHz and 640 kHz. A 12.8 MHz oscillation does not excite a 148 kHz resonance directly. But the restrike train is a burst of several hundred steps spread over a fraction of a second, and the spectral content of that burst envelope — as distinct from the content of any single front — has significant energy in the 100 to 700 kHz band. This is the mechanism by which repeated disconnector operation can excite a part-winding resonance that no single event would reach, and it is why restrike count matters as much as restrike magnitude. The second is surge transfer to the 138 kV winding. The high-frequency capacitive coupling between windings transferred 0.11 pu of the 345 kV side VFTO to the 138 kV terminals, or 76 kV on a 138 kV base — not a dielectric concern on its own, but relevant because the 138 kV side connects to cable and the transferred front propagates further. Transformer Winding Stress Summary

The ratio does not improve with mitigation because it is a property of the winding, not of the stress. Only the magnitude improves — which is the whole point of mitigating at the source.

7.5 Transient enclosure voltage results

The internal traveling wave is confined by the coaxial geometry, and at every discontinuity in that geometry a fraction of it couples to the outer surface of the enclosure. The enclosure then behaves as a transmission line with respect to remote earth, with a computed surge impedance of 130 Ω, and supports its own oscillation determined by its length and its terminations. TEV Results, Pre-Mitigation

The 6.1 MHz dominant frequency corresponds to a quarter-wave resonance on 40 feet of enclosure, which is the run from the bushing flange to the nearest effective bond. That phrase — nearest effective bond — is the finding. There were bonding straps far closer than 40 feet. They were not effective. A bonding strap is an inductor. At 0.32 µH per foot, one of the surveyed 10.5-foot straps of 4/0 AWG copper presents 3.36 µH, and at 6.1 MHz its impedance magnitude is 129 Ω. The enclosure surge impedance is 130 Ω. The strap is, at this frequency, indistinguishable from no connection at all. At 60 Hz the same strap has an impedance of 1.3 milliohms and is an excellent ground. This is the single most important practical insight in the study and the one the Owner's staff found most counterintuitive: the station's grounding system was correctly designed, correctly installed, correctly tested, entirely compliant with IEEE Std 80, and completely irrelevant to the phenomenon causing the shocks. The remedy follows from the same arithmetic. Impedance is proportional to length, so halving the strap halves the impedance. Reducing the surveyed straps to under 2 feet and replacing round conductor with wide copper sheet — which has lower inductance per unit length because of its larger surface — brought the effective bond impedance at 6.1 MHz from 129 Ω to 21 Ω. Combined with additional bonds so that no enclosure run exceeded 10 feet between effective bonds, the resonant length dropped and the resonant frequency rose out of the band where the internal coupling has significant energy. Modeled TEV at the governing bushing flange fell to 1.9 kV.

7.6 Coupling into secondary circuits

Secondary circuits are exposed by three mechanisms, and the study found all three active. Conductive coupling occurs where a secondary cable's shield or a device's ground reference is bonded to an enclosure that is itself at 8.4 kV with respect to remote earth. The full TEV appears as a common-mode voltage between that reference and the reference at the other end of the cable. Inductive coupling occurs because the ground strap and enclosure currents are large and fast. Simulated peak current in the governing bushing bonding strap was 683 A with a 4.6 ns front, giving a di/dt of the order of 10¹¹ A/s. A cable routed parallel to that strap for 16 feet at 12 inches separation has a mutual inductance of roughly 0.9 µH, and the induced voltage follows directly. Radiative coupling occurs because a 6.1 MHz oscillation on a 40-foot enclosure is an efficient antenna. Field strengths in the immediate vicinity of a bushing flange during a restrike are substantial, and any cable within a few feet is in the near field. Secondary Circuit Induced Voltage

The measured 3.1 kV must be compared against what the relays were actually qualified to. IEEE Std C37.90.1 specifies a 2.5 kV crest oscillatory surge withstand test at 1 MHz, and a 4 kV fast transient test. The measured stress exceeded the oscillatory crest by 24 percent and did so at 5 to 14 MHz, five to fourteen times the test frequency, where cable and equipment coupling behavior is entirely different. The IEC 61000-4-18 damped oscillatory wave test extends to 100 kHz and 1 MHz and likewise does not reach this band; IEC 61000-4-4 fast transient bursts have fast fronts but a 5 kHz repetition structure and a very different coupling path. The honest conclusion, and the one the study stated, is that no immunity test in either standards family qualifies a relay against the stress present in this station, and the correct response is therefore to reduce the stress rather than to demand a better-qualified relay. The single-end versus double-end shield bonding result was the one that changed practice. Bonding a shield at one end only prevents circulating power-frequency current and is standard practice in many US utilities for that reason. It also leaves the shield electrically open at high frequency at one end, so it cannot carry the return current that would otherwise cancel the induced core voltage. Bonding both ends reduced the induced voltage from 2.2 kV to 640 V. The circulating-current objection is addressed by the parallel ground continuity conductor, which carries the low-frequency current the shield would otherwise carry.

7.7 Circuit breaker switching and TRV verification

The GIS breakers were verified against IEEE Std C37.06 for terminal fault, short-line fault and out-of-phase duties. GIS is generally favorable for TRV because the bus capacitance reduces the rate of rise, and that was confirmed here. TRV Verification Against IEEE Std C37.06

All duties passed with margin. The short-line fault case is the closest, as it usually is, because the line-side sawtooth component is set by the overhead line surge impedance and the fault current, neither of which the GIS capacitance influences much. No mitigation was required. This part of the study is reported because a complete transient study of a station should verify TRV, and because a negative result here would have changed the mitigation package substantially.

7.8 Energization and lightning ingress

Transformer energization produced inrush with a 42 percent second-harmonic content and a maximum sympathetic interaction with the adjacent bank of 0.14 pu current, neither presenting a dielectric or protection concern; the existing second-harmonic restraint was confirmed adequate. Cable energization on the 138 kV side produced a maximum overvoltage of 1.71 pu, within the arrester protective level with a 38 percent margin. Lightning ingress was evaluated at the overhead-to-GIS interface, which is the only path by which a lightning surge can reach the GIS. A shielding failure stroke of 12 kA and a backflashover-driven surge of 68 kA were injected at the transition structure. With the existing line-entrance arresters the maximum voltage inside the GIS was 743 kV against the 1050 kV BIL, a protective margin of 41 percent, satisfying the 20 percent minimum of IEEE Std 1313.2. Lightning is a slower and, in this station, less demanding stress than VFTO in rate of rise, though larger in magnitude. That comparison — a 68 kA lightning surge producing a gentler front than a routine disconnector operation — is the clearest single illustration of why the study was necessary.

8. Sensitivity and Scenario Analysis

Six parameters were varied to establish which conclusions were robust and which depended on assumptions the study could not fully verify. Sensitivity Results

Three conclusions follow. First, VFTO magnitude is dominated by trapped charge and almost indifferent to the spark model parameters; the spark parameters set the rise time, not the peak. This is fortunate, because trapped charge was measured and the spark parameters were assumed. Second, the frequency is set by geometry and terminal capacitance, so it can be predicted at design stage from drawings alone — which is the basis of the specification language delivered in Section 11. Third, TEV is overwhelmingly a function of bonding strap length and essentially nothing else within the ranges tested, which is why the mitigation is cheap and why the problem was avoidable. What did not matter: GIS section losses, changed by a factor of ten with less than 1 percent effect on peak; the number of bays energized, which shifted frequency by under 4 percent; and the exact overhead line length beyond the first mile, since nothing returns from further away inside the 20 µs window.

9. Findings and Root Cause Assessment

Findings Register

The root cause has one root and several branches. The root is that the GIS was procured, designed, installed and commissioned by parties none of whom treated very fast transients as a design input. The disconnectors were supplied without damping resistors because none were specified; the bonding was designed to a 60 Hz grounding standard and satisfied it; the secondary cabling was routed for convenience and shielded to a practice that predates numerical relays; and the commissioning tests, all of which passed, measured nothing above a few kilohertz. The branches are the specific accumulations. F1 arose from field routing decisions taken during installation, where a strap was lengthened to clear an obstruction with no recognition that length was the parameter that mattered. F5 and F9 arose from cable tray congestion. F3 arose from a bus length that happened to place a quarter-wave resonance at the frequency the disconnector excites, an outcome of general arrangement drafting rather than of any electrical decision. F10 is the reason all of the above would have repeated in the planned extension. It is worth recording what was not the cause. The relays were sound. The grounding grid was sound. The SF6 system, the disconnector mechanisms and the breakers were all within specification and correctly maintained. Nothing was broken. The station was assembled from compliant parts into a configuration whose emergent behavior no party had analyzed.

10. Mitigation Options and Recommendations

Mitigation Options Comparison

Each option was modeled in the same framework, and three deserve comment because their intuitive appeal exceeds their performance. Surge arresters at the transformer terminal are the instinctive answer and are largely ineffective against VFTO. A metal-oxide arrester's protective action depends on the block conducting, and the arrester has its own inductance — lead inductance plus the internal column — which at a 4.6 ns front dominates its behavior. The arrester clips the peak of a slow surge; against a nanosecond front it inserts an inductive voltage that partly defeats its own purpose, and the separation distance between arrester and transformer, trivial at lightning frequencies, is a significant fraction of a wavelength at 12.8 MHz. Arresters remain necessary for lightning and switching duty and were retained. They are not a VFTO mitigation. Ferrite rings fitted around the enclosure at discontinuities add high-frequency series impedance to the enclosure current path and are genuinely useful, cheap and easy to retrofit. Their limitation is saturation at the current levels present here and their negligible effect on the internal VFTO. They were adopted as a supplement, not a primary measure. Damping resistors are the only effective VFTO mitigation. A resistor of the order of the bus surge impedance, in series with an auxiliary contact that makes before the main contact, damps the traveling wave at the instant it is launched. At 500 Ω the modeled VFTO fell from 2.34 to 1.62 pu, the ringing decayed in under 2 µs instead of 14 µs, and the winding stress fell in proportion. The cost is high because the disconnectors must be replaced rather than modified and the work requires a bay outage. The recommendation. A four-part package. First, replace the four disconnectors adjacent to the two transformer bays with damping-resistor-equipped units, during scheduled bay outages, one bay at a time. The transformer bays were selected because they carry the highest VFTO and the most valuable and least testable asset. The remaining disconnectors were left unmodified, a decision the study justified explicitly: with the enclosure and secondary mitigations in place, their VFTO of up to 2.11 pu stresses GIS equipment whose gas-insulated dielectric is far more tolerant of fast fronts than an oil-paper winding, and the marginal risk reduction did not justify eight further bay outages. Second, rebuild the enclosure bonding: no strap longer than 2 feet, wide copper sheet in place of round conductor, additional bonds so that no enclosure run exceeds 10 feet between effective bonds, and an equipotential mesh under all accessible platforms and walkways bonded to the enclosure at both ends of each run. This is the cheapest element of the package and delivers the personnel safety outcome outright. Third, reroute and re-terminate the affected secondary cabling: shielded cable throughout, shields bonded at both ends per IEEE Std 525, a parallel ground continuity conductor along every cable route between the GIS and the relay building, minimum 3 feet separation from enclosure discontinuities, and physical separation of AC and DC circuits into different trays. Fourth, retain the operational restriction on disconnector operation with personnel in the affected bay until the bonding work is complete, then withdraw it.

11. Implementation Support and Field Validation

The field measurement campaign served two purposes: it validated the model before the recommendation was issued, and it verified the result after implementation. Enclosure voltage was measured with capacitive dividers of 1000:1 ratio and 200 MHz bandwidth, mounted directly on the enclosure surface at six locations with the low-voltage arm referenced through a short coaxial run to a local reference plane. Currents in bonding straps were measured with wideband current transformers covering 100 Hz to 100 MHz. Internal VFTO could not be measured directly — there is no access to the conductor — so it was inferred at one location from a capacitive divider installed at a spare flange during the outage, a technique that gives a good waveform and frequency but a ratio uncertain by roughly 10 percent, which the study states as a limitation on that one quantity. Acquisition used an eight-channel digitizer at 2.5 GS/s and 500 MHz analog bandwidth, with all signal runs double-shielded and the instrument battery-powered inside a shielded enclosure to keep the measurement system from measuring itself. Eighteen disconnector operations were recorded pre-mitigation and twelve post-mitigation. Field Validation Against Simulation

Every quantity fell inside the acceptance criteria of Section 6.4. The consistent pattern of simulated frequencies slightly above measured and simulated decay slower than measured is the expected signature of a lossless model and confirms the direction of the conservatism rather than undermining it. Post-implementation measurement confirmed TEV at the governing bushing flange of 1.9 kV against a modeled 1.8 kV, and secondary induced voltage of 280 V against a modeled 260 V. Implementation support included the bonding design drawings, the damping resistor technical specification, cable routing drawings, witness of the first bay retrofit, and the specification language added to the Owner's GIS procurement standard. That language requires the supplier to state guaranteed VFTO limits at defined locations, to provide the section geometry needed for an independent VFTO study, to offer damping-resistor disconnectors as a priced option, to state maximum bonding strap length as a design constraint rather than a field decision, and to demonstrate TEV below 2.0 kV at all accessible surfaces during type or site testing.

12. Results and Value Delivered

Outcomes Scorecard

Twenty-two months after completion of the retrofit the station has recorded no protection misoperation attributable to switching, no shock report, and no change in transformer dissolved gas trend. The validated model was transferred to the Owner and has since been used to evaluate the two-bay extension at design stage, where a bus arrangement placing a quarter-wave resonance close to the disconnector excitation band was identified on drawings and changed before procurement — at no cost, which is the difference between doing this analysis at design stage and doing it after commissioning.

13. Lessons Learned and Engineering Insights

A grounding system can be fully compliant and completely ineffective. The station's grid met IEEE Std 80 and its bonding met the installation specification. At 6.1 MHz a 10.5-foot bonding strap is a 129 Ω impedance against a 130 Ω enclosure — electrically, not a connection. Grounding adequacy is frequency-dependent, and no power-frequency test will reveal a high-frequency bonding failure. Where GIS is involved, strap length is a dielectric and safety parameter and belongs on the drawing with a maximum dimension, not left to field judgment. The BIL test does not cover the stress that dominates GIS. A 1050 kV BIL demonstrated with a 1.2/50 µs impulse says nothing useful about a 4.6 ns front applied several hundred times per switching operation. Reporting a comfortable margin against BIL for a very-fast-front stress is a category error. IEEE Std 1313.1 defines a separate very-fast-front class precisely because the standard waveforms do not reach it. Single-end shield bonding is a low-frequency practice applied in a high-frequency environment. The rule exists to prevent circulating power-frequency current, a real problem with a better solution — a parallel ground continuity conductor. Leaving a shield open at one end removes its ability to carry the high-frequency return current that would cancel the induced core voltage. Here the difference was 2.2 kV against 640 V from that decision alone. Model bandwidth is a scoping decision, and scoping decisions hide phenomena. The procurement-stage study was competent and correct for the questions it asked. Its time step made VFTO invisible, not small. When a client says a phenomenon was studied and not found, the first question is what time step and what topology, not what result. Mitigate at the source; downstream protection does not work against nanosecond fronts. Surge arresters, the instinctive remedy, are largely ineffective here because their own inductance and separation distance dominate at 12.8 MHz. Damping resistors work because they act at the instant the wave is launched, before any propagation has occurred. As a general principle for very fast transients, every meter between the mitigation and the source costs 3.4 ns of ineffectiveness. The cheapest fix addressed the most serious risk. The bonding rebuild was the least expensive element of the package and it delivered the personnel safety outcome and most of the interference reduction. The expensive element, disconnector replacement, addressed an asset risk that was real but slower-acting. Sequencing mitigation by cost-to-benefit rather than by severity of the headline finding got the shock hazard eliminated months before the retrofit outages began.

14. Keentel Capability Summary

• Very fast transient overvoltage (VFTO) studies for gas-insulated substations at all US transmission voltage classes
• Transient enclosure voltage (TEV) and transient ground potential rise assessment, including personnel exposure evaluation
• Nanosecond-resolution EMT modeling in PSCAD/EMTDC and EMTP, with distributed-parameter GIS section models and spark models for disconnector arcs
• High-frequency transformer modeling from on-site terminal impedance measurement, including initial winding distribution and part-winding resonance screening
• Electromagnetic compatibility assessment of substation secondary systems against IEEE Std C37.90.1 and the IEC 61000-4 series
• Substation grounding and bonding assessment at high frequency, including equipotential mesh design and bonding impedance verification
• Transient recovery voltage verification against IEEE Std C37.06 and IEEE Std C37.011, including terminal fault, short-line fault and out-of-phase duties
• Insulation coordination to IEEE Std 1313.1 and 1313.2 across temporary, switching, lightning and very-fast-front classes
• High-bandwidth field measurement campaigns using capacitive dividers, wideband current transformers and gigasample acquisition, with formal model validation criteria
• Preparation of GIS procurement specification language covering VFTO limits, TEV limits, damping resistor options and bonding design constraints

15. Frequently Asked Questions

16. Glossary of Terms and Abbreviations

17. Confidentiality and Use Statement

This case study has been prepared for informational purposes. All client identities, project locations, contract details, and proprietary data have been withheld or generalized. Technical parameters, study results, and figures presented are representative of work performed by Keentel Engineering Solutions and have been adapted so that no individual project, owner, or facility can be identified. Nothing in this document constitutes a design recommendation for any specific installation. Any reuse of the methodologies described requires project-specific engineering analysis by a qualified professional engineer.