Insulation Coordination and Surge Protection Study for a High-Voltage Substation

1. Executive Summary

Keentel Engineering Solutions was retained by the Owner of a 500/230 kV bulk transmission station to determine why the station had been consuming surge arresters at a rate no one could explain, and whether the original insulation coordination scheme remained adequate. Over the six operating years preceding Month 0 the station had suffered seven metal-oxide surge arrester (MOSA) failures — five on the 500 kV line entrances, two on the 230 kV bus — and one 230 kV transformer bushing failure that released oil and removed a 500 MVA autotransformer bank from service for eleven weeks. Asset management had replaced arresters like for like each time. The failures continued. Answering the question required rebuilding the insulation coordination from first principles — the full chain from electrical stress through insulation strength to protective margin — rather than auditing the original calculation, a twelve-page hand computation to a superseded practice containing no statistical switching study, no backflashover calculation, no station shielding analysis and no temporary overvoltage check. The study was executed in PSCAD/EMTDC with supporting work in EMTP, ASPEN OneLiner, CDEGS and Python, and covered all three overvoltage classes recognized by IEEE Std 1313.1. The deterministic protective-margin method served as the primary screen; the statistical risk-of-failure method resolved the cases that screen flagged. Three independent root causes emerged, all products of change accumulated over the asset's life rather than original design error. First, the arrester nominally protecting the Bank 2 autotransformer 230 kV bushings sat 151 feet of conductor away following a bay reconfiguration in which bushing-mounted arresters had been removed and never reinstated; the travel-time contribution drove the lightning protective margin to negative 8.1 percent. Second, retirement of nearby generation had weakened the station's zero-sequence source, raising the earth fault factor from the 1.25 assumed originally to 1.35 as found, pushing the TOV duty to 1.42 per unit for 1.1 seconds — onto the installed arresters' TOV withstand curve with no margin. Third, a breaker retrofit had replaced pre-insertion-resistor-equipped 500 kV breakers with resistor-less units, so that reclosing onto trapped charge produced a statistical 2 percent switching overvoltage of 2.9 per unit and an arrester energy duty of 4.9 MJ against a 3.09 MJ capability. The recommended program — re-specified arresters, arrester relocation, controlled closing retrofit, counterpoise on the approach towers, a 230 kV grounding bank and a backup clearing time reduction — restored every protective margin above 52 percent, cut the computed risk of insulation failure per switching operation from 4.6 × 10⁻³ to 2.4 × 10⁻⁵, and eliminated the failure mode. No arrester or bushing failure has occurred in the twenty-eight months since implementation. 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 terminates three 500 kV lines in a breaker-and-a-half arrangement and steps a substantial block of that transfer down to a 230 kV network through two autotransformer banks, each three single-phase units of 167 MVA with a fourth held as site spare and a buried 13.8 kV delta tertiary carrying station service. The 230 kV yard is also breaker-and-a-half and serves four lines. The station is a critical element of the Owner's transfer path and is not redundant in any operationally meaningful sense. The failure history drove the study; the commercial stake sat behind it. The eleven-week bushing outage was absorbed without a reliability event only because it fell in a light-load season. The same failure on the other bank during heavy transfer would have produced a different outcome, and a failure damaging the transformer active part rather than the bushing alone would have introduced a fourteen-month replacement lead time against a spare policy that covers one phase, not three. The deeper driver was change. Three material changes had occurred since the original design, none of which triggered an review because none was recognized as touching it. A 500 kV breaker retrofit replaced the original units, which carried pre-insertion resistors, with modern resistor-less breakers of higher interrupting rating — a straightforward duty upgrade that silently removed the station's principal switching surge control. A bay reconfiguration on the 230 kV side relocated the Bank 2 high-side arresters from the bushing risers to a dead-end structure at the far end of the bus section. And a block of generation electrically close to the station retired, weakening the local zero-sequence source and changing the earth fault factor at both buses. Each change was defensible in isolation. Their combined effect on insulation coordination had never been evaluated, because insulation coordination is not a routine deliverable of a breaker retrofit, a bay reconfiguration, or a generation retirement study. That is the general lesson the engagement returned, and it is why the study was scoped as a rebuild rather than an audit.

3. Study Objectives and Scope of Work

1. Establish the maximum credible temporary overvoltage at the 500 kV, 230 kV and 13.8 kV buses under load rejection, ground fault, resonance and ferroresonance, and verify the installed arresters against their TOV withstand characteristic.
2. Determine, by statistical switching study, the distribution of switching overvoltage magnitude for line energization, reclosing onto trapped charge, shunt reactor switching and capacitor bank switching, and verify switching impulse withstand levels.
3. Determine station lightning performance, treating shielding failure and backflashover as distinct mechanisms, and verify lightning impulse withstand levels including the effects of arrester lead length and separation distance.
4. Compute protective margins for every major item of plant against the 20 percent minimum recommended by IEEE Std 1313.2, resolving marginal deterministic results by statistical risk-of-failure analysis.
5. Establish root cause of the historical failures to a standard of evidence sufficient to support a capital request.
6. Compare mitigation options on technical merit, order-of-magnitude cost, schedule and residual risk, and issue a recommendation.
7. Provide equipment specification support, relocation drawings and field validation of the implemented scheme.

Scope and Deliverables

Also excluded and documented as such: transient recovery voltage studies, station ground grid fault performance, arc flash, and evaluation of autotransformer internal winding stress beyond the terminal surge transfer check of Section 7.6. The Owner accepted these exclusions in writing.

4. Regulatory and Standards Basis

The regulatory driver is the obligation on the Transmission Owner to maintain facility ratings and equipment performance consistent with planning assumptions. NERC FAC-008 requires facility ratings consistent with equipment ratings; plant whose insulation is stressed beyond its rated withstand is not being operated within its rating. NERC TPL-001 performance requirements assume equipment survives the events studied, an assumption repeated dielectric failure undermines. The backup clearing time reduction recommended in Section 10 was coordinated against NERC PRC-027 obligations, and the Owner's protection group confirmed selectivity was preserved. Standards and Regulatory Register

Where a US convention differs from the IEC equivalent, the US convention governs. Insulation levels are expressed as BIL and BSL from the preferred-level tables of IEEE Std C57.12.00 and C37.06, not as IEC values assigned against a highest-voltage-for-equipment ladder. The IEC comparison is noted where it changes the numerical answer, and nowhere else.

5. System Modeling and Data Development

5.1 Data sources and gap closure

Model Data Sources and Gap Closure

The geometry survey was the study's most valuable data activity. As-built drawings showed the Bank 2 230 kV arresters on the bushing risers. They had not been there for years, no drawing had been revised, and no one still at the station remembered the change. The survey found separation distances between 15 feet and 151 feet against a drawing set implying a maximum of 62 feet.

5.2 Model construction

The PSCAD/EMTDC model was built at three levels of detail, selected by phenomenon. For lightning, the last five towers of each incoming line were represented individually: each tower as a multi-section distributed surge impedance of 180 Ω with 0.15 µs travel time for the 148-foot structure, each insulator string as a leader-progression flashover element, and each footing as a current-dependent nonlinear resistance using the measured low-current value and a 400 kV/m soil ionization gradient. Beyond the fifth tower, ten further spans of frequency-dependent phase-domain line were included, terminated in a matched 380 Ω resistance so that reflections from the artificial termination could not return within the 30 µs window; the two-way travel time to termination is 39.6 µs, confirming the window is clean. For switching, the whole 214-mile incoming 500 kV line was modeled with a frequency-dependent phase-domain formulation together with its 200 MVAr shunt reactor, the remote source equivalent and full station bus geometry. For TOV and resonance the model added transformer saturation and extended to the second tier of buses, reduced beyond that to a Thevenin equivalent validated against a harmonic impedance scan of the full case. Line model selection was explicit. The Bergeron model, a constant-parameter distributed model evaluated at a single frequency, is valid where one frequency dominates and where damping and waveshape distortion accuracy are not the answer sought — steady state, and rough switching estimates on short lines. It is invalid for lightning, where the front contains energy from tens of kilohertz to megahertz and where ground-return impedance and skin effect vary by orders of magnitude across that band. It was also inadequate for the switching magnitudes needed here, because the modal attenuation limiting the open-end surge is frequency dependent. Frequency-dependent models were therefore used for all lines of interest, with Bergeron retained only for the network equivalent beyond 20 miles. The autotransformer was represented three ways: a saturable model with a 1.15 pu knee and measured air-core reactance for TOV; the conventional leakage-impedance representation for switching; and for lightning and surge transfer a measured capacitance network with CH of 4,200 pF, CHL of 3,100 pF and CL of 6,800 pF, plus series winding capacitance so the model reproduced terminal impedance to 1 MHz. A lumped ground capacitance alone would have sufficed for terminal voltage but would have given a meaningless tertiary transfer result. Arresters used the IEEE frequency-dependent arrester model — two nonlinear resistance branches separated by an R-L filter, parameterized from 8/20 µs and steep-front discharge data and physical arrester height. At switching frequencies the filter is transparent and the model reduces to a single V-I characteristic, which is correct. For lightning the filter reproduces the dynamic overshoot a static curve cannot, and this matters: the dynamic characteristic gives a discharge voltage several percent above the static curve on a fast front, so a study ignoring it reports margins that are optimistic in exactly the cases where margin is scarce. Corona was represented on the incoming spans with a piecewise charge-voltage model and a 1,400 kV onset for the 500 kV bundle. Corona limits how steep an incoming surge can be by the time it reaches the station; omitting it makes the wave steeper and the study more pessimistic, but also less able to distinguish a real problem from a modeling artifact. This study needed that distinction. Time steps were 2 ns for lightning over a 30 µs window, 10 µs for switching over 500 ms, and 20 µs for TOV over 6 s. The lightning step was verified by halving it to 1 ns, changing peak terminal voltages by less than 0.4 percent.

5.3 Model validation

Model Validation and Benchmarking

The tool-diversity cross-check was performed because the switching result was going to drive a seven-figure capital recommendation, and a single-tool result carrying that weight is not defensible. An independent engineer built the case in EMTP from the same data sheet without sight of the PSCAD results.

6. Study Methodology and Assumptions

6.1 The insulation coordination chain

IEEE Std 1313.1 frames insulation coordination as three steps: determine the stress, determine the strength, establish the relationship between them. Stresses are classified by waveshape and duration. Temporary overvoltages are at or near power frequency, last from a cycle to minutes, and are characterized by magnitude and duration. Switching overvoltages have fronts of tens to thousands of microseconds and are statistical by nature, because magnitude depends on the point on wave at which each breaker pole closes. Lightning overvoltages have fronts of a fraction of a microsecond to a few microseconds, are dominated by traveling-wave behavior, and are statistical in a different sense — magnitude depends on the stroke current distribution and on which of two distinct mechanisms, shielding failure or backflashover, produced them. Strengths are the standard withstand levels: BIL for the lightning impulse, BSL for the switching impulse, chopped wave withstand for the truncated lightning impulse, and power frequency withstand for TOV. Self-restoring insulation — air gaps and the external surfaces of bushings and insulators — has a strength described statistically by a critical flashover voltage (CFO) and a standard deviation. Non-self-restoring insulation — transformer and instrument transformer internal insulation, cable, bushing internals — fails permanently, is treated deterministically, and calls for stress kept comfortably below withstand rather than a calculated risk. That distinction drove method selection here: the 500 kV air clearance switching performance was resolved statistically, while the transformer bushing problem that started the engagement was resolved deterministically, because there is no acceptable risk of bushing failure tradeable against a modest capital saving.

6.2 Deterministic and statistical methods

The deterministic method computes protective margin, percent = (withstand voltage / protective level − 1) × 100, with IEEE Std 1313.2 recommending a 20 percent minimum. The protective level is the voltage appearing at the terminals of the protected equipment, not the arrester's catalog discharge voltage. That distinction is the subject of Section 7.3 and is where this station's problem lived. The statistical method exists in two forms. The two-parameter method characterizes stress by its 2 percent value E2 — the magnitude exceeded in 2 percent of operations — and strength by the statistical withstand voltage V3, defined as CFO minus three standard deviations, at which flashover probability is about 0.13 percent. The statistical safety factor is γ = V3 / E2, with 1.15 conventional US practice for self-restoring insulation. It is a screening tool: it collapses two distributions into one point each and says nothing about the shape of either tail. The full risk-of-failure method integrates the overvoltage probability density against the insulation cumulative discharge probability, R = ∫ f(V) · P(V) dV, evaluated per operation and multiplied by the number of insulation elements in parallel exposure and the annual operation count. This study screened with the two-parameter method and resolved with the full integration, because two switching cases produced γ values close enough to criterion that a decision on the screening value alone would have been arbitrary.

6.3 Tools

Tools and Selection Rationale

6.4 Where US practice diverges from IEC 60071

IEEE assigns discrete preferred insulation levels — BIL 1800 kV and BSL 1300 kV for 550 kV maximum equipment, BIL 900 kV and BSL 745 kV for 242 kV maximum equipment — and asks the engineer to demonstrate margin against them directly. IEC 60071-2 builds a four-step chain from the representative overvoltage through a coordination withstand voltage (coordination factor Kc), to a required withstand voltage (atmospheric correction Ka and safety factor Ks), to a standard rated withstand voltage from a table. The IEC safety factor Ks of 1.15 for internal and 1.05 for external insulation performs roughly the work of the IEEE 20 percent margin, but is applied at a different point against a different reference. Three divergences mattered here. The standard levels differ: IEC assigns 1550 or 1675 kV to 550 kV equipment where IEEE assigns 1800 kV, so an IEC-derived margin cannot be read across. IEC permits, for equipment at or below 245 kV, that switching impulse withstand be demonstrated by conversion from the lightning impulse test; US practice at 242 kV assigns and where necessary tests a BSL. And IEC treats atmospheric correction as an explicit factor in the required withstand chain, while US practice more often applies altitude correction to available strength. This study applied correction to strength and stated the choice so the numbers can be audited.

6.5 Case matrix and acceptance criteria

Case and Scenario Matrix

Acceptance Criteria

7. Analysis and Results

7.1 Temporary overvoltage

TOV was analyzed first, because TOV determines arrester rating, rating determines discharge voltage, and discharge voltage determines every protective margin downstream. An arrester chosen too low for the TOV fails thermally; one chosen too high protects poorly. The scheme is anchored here. Maximum continuous operating voltage at the 500 kV bus is 550 kV line to line per ANSI C84.1, giving 317.5 kV line to ground, against which the installed MCOV of 318 kV is correctly selected. At 230 kV the figures are 242 kV, 139.7 kV and an installed MCOV of 144 kV. Both selections were sound. The problem was not MCOV; it was what happens above MCOV. Ground fault overvoltage. During a single line to ground fault the unfaulted phases rise. How far is governed by the coefficient of grounding — the ratio of the highest power-frequency line-to-ground voltage on an unfaulted phase during the fault to the line-to-line voltage with the fault removed — and equivalently by the earth fault factor, √3 times that coefficient, referenced to pre-fault line-to-ground voltage. Both are functions of X0/X1 and R0/X1 at the fault location. The original design assumed X0/X1 of 3.0 and R0/X1 of 1.0, the classic effectively-grounded boundary, giving an earth fault factor of 1.25 and a coefficient of grounding of 0.72. The as-found sequence impedances told a different story. With the nearby generation retired, its grounded-wye-delta step-up banks no longer contributed zero-sequence current at this station. Recomputation gave X0/X1 of 4.6 and R0/X1 of 1.4 at the 500 kV bus, an earth fault factor of 1.35. Applied to a pre-fault operating voltage of 1.05 pu, the duty is 1.42 pu of nominal line-to-ground voltage, or 410 kV rms. Duration is set by protection: primary clearing gives 5 cycles, but the governing case is a stuck breaker cleared by breaker failure backup in 1.1 seconds. Against the arrester, 410 kV is 1.29 times the 318 kV MCOV for 1.1 seconds. The installed arresters' published TOV capability with prior energy is 1.30 pu at 1.0 s and 1.27 pu at 3 s, interpolating to about 1.295 pu at 1.1 s. Duty and capability are the same number to within half a percent. The station had been operating with essentially zero arrester TOV margin, and every stuck-breaker ground fault was a coin toss. Load rejection and Ferranti rise. Full rejection leaves the 214-mile incoming line lightly loaded and the source behind it briefly overspeeding. The distributed-parameter open-end rise for that length at 60 Hz is 1.103 pu with the 200 MVAr shunt reactor out of service, confirmed analytically. With source regulation and overspeed the simulated bus TOV reached 1.31 pu, decaying over about 3 seconds. Against the arrester that is 1.19 times MCOV for 3 s versus a capability of 1.27 pu — a ratio of 1.067, below the 1.10 criterion though not by itself a failure mechanism. Resonance, ferroresonance and saturation. A harmonic impedance scan of the 230 kV bus with one bank energized showed a parallel resonance near 240 Hz formed between line capacitance and transformer and source inductance. Energizing the second bank against the first produces a sympathetic inrush rich in even harmonics; the fourth-harmonic component is amplified by 2.1, sustaining 1.18 pu for 4.2 seconds before the inrush decays. This is within arrester capability but was documented as an operating consideration. Ferroresonance was more serious. The 500 kV breakers carry grading capacitors of about 1,200 pF per break, four breaks across a bay. With a bus section de-energized by opening its disconnects while adjacent bay breakers remain closed, the bus voltage transformer is fed through that capacitive path into a saturable magnetic circuit — the textbook ferroresonant arrangement. Of twelve switching orders simulated, two produced sustained subharmonic ferroresonance near 20 Hz at 2.1 pu peak, persisting indefinitely with the VT core deep in saturation. The mode had never been observed because those two orders are not the ones normally used; but nothing prevented an operator from using them. Temporary Overvoltage Results

7.2 Switching surges

Switching overvoltages are statistical because breaker poles do not close at a defined instant. Mechanical scatter of one to two milliseconds translates directly into point-on-wave scatter, and surge magnitude varies widely from operation to operation. A single deterministic simulation of line energization is worth very little. The study used 400 randomized runs per case, with pole closing instants drawn from a Gaussian distribution of 1.2 ms standard deviation about the commanded instant plus a systematic pole span representative of the installed breakers, and built a cumulative distribution from which E2 was extracted. All statistical values below are prospective values, computed with station arresters removed from the model. This follows the two-step logic of IEEE Std 1313.2: the prospective stress establishes what the arrester must control and therefore its energy duty, while the arrester-limited stress establishes margin against BSL. Reporting only the arrester-limited value hides the arrester duty, which here was the failure mechanism. The worst case is reclosing onto trapped charge. When a breaker interrupts a line, the line capacitance retains charge at whatever instantaneous voltage existed at current zero, up to 1.0 pu of crest. If the breaker recloses while that charge remains and the source is at the opposite polarity peak, the step across the line is 2.0 pu, and the traveling wave doubles at the open receiving end to approach 3.0 pu. Everything about switching surge control is an attempt to prevent that coincidence. This line has a 200 MVAr shunt reactor, giving about 40 percent shunt compensation. That is normally beneficial but it changes trapped charge behavior in a way often missed. Rather than decaying slowly through insulation leakage over minutes, the trapped charge oscillates with the reactor at the line's natural frequency of 37.9 Hz. Because that is not 60 Hz, line-side and source voltages beat against each other with a period near 45 ms. A fixed 0.5 second reclosing dead time therefore does not reliably reclose at a favorable instant — it recloses at whatever phase of the beat falls there, and across a population of operations it will sometimes find the worst instant available. With the original pre-insertion-resistor breakers this was controlled: a 400 Ω resistor inserted for 8 to 10 ms damps the traveling wave during the interval when it would otherwise double. The retrofit removed the resistors. Statistical Switching Overvoltage Results, 500 kV

The per-unit base is the crest line-to-ground voltage at maximum system voltage, 449.1 kV. The as-found reclosing case of 2.90 pu equals 1,302 kV against a BSL of 1,300 kV. With arresters in service the terminal stress is clipped well below that, so BSL is not directly violated — but the arrester is doing all the work, and Section 7.4 shows what that costs it. The mitigation choice was between pre-insertion resistors and controlled (synchronous) closing. Resistors are proven and independent of measurement, but require a breaker with the resistor mechanism, which the retrofitted units cannot be given without replacing the interrupter assemblies. Controlled closing commands each pole to close at a target instant referenced to measured line-side and source-side voltages, and on a compensated line can track the beat and target minimum voltage across the contacts. Its weakness is dependence on measurement, controller and repeatable mechanical operating time, and it degrades with breaker wear. The study modeled a residual scatter of 0.6 ms, a realistic in-service rather than factory value, obtaining E2 of 1.90 pu. The 230 kV switching cases were not governing: capacitor bank energization gave E2 of 2.4 pu on a 197.6 kV base, or 474 kV, a margin of 57 percent against the 745 kV BSL.

7.3 Lightning surges and the separation distance problem

Lightning reaches station equipment by two mechanisms that must be analyzed separately because they produce different waveshapes and magnitudes. Shielding failure occurs when a stroke bypasses shield wires or masts and terminates on a phase conductor; the current is limited, because a large stroke has a larger striking distance and is intercepted, so shielding failures deliver modest currents with steep fronts. Backflashover occurs when a stroke terminates on shield wire or tower top, raises tower and footing potential above conductor potential, and flashes the insulator string from grounded steel to conductor; it requires a large current and a high footing impedance and dominates on well-shielded high-voltage lines. Station shielding was assessed to IEEE Std 998 using the rolling sphere method, with sphere radius set by the striking distance for the maximum allowable stroke current, Is = 2.2 × BIL / Zs, where Zs is the bus conductor surge impedance of 400 Ω. This gives 9.9 kA at 500 kV with BIL 1800 kV and 4.95 kA at 230 kV with BIL 900 kV. The corresponding striking distances from S = 8 k I^0.65, with S in meters and k = 1 for wires, are 35.5 and 22.6 m, or 116 and 74 feet. The 230 kV yard therefore requires a tighter shield than the 500 kV yard, the opposite of intuition. Rolling the 74-foot sphere over the 230 kV yard found two penetration zones: a section of bus where the shield wire span sagged below the required height, and the Bank 2 230 kV riser conductors, outboard of the shield envelope. The 116-foot sphere found no penetration in the 500 kV yard. The 230 kV exposure was real but small — computed direct stroke incidence of 0.011 strokes per year — and was not the cause of the observed failures. It was corrected anyway with two 80-foot masts. Backflashover performance of the approach spans was the significant lightning finding. Footing resistance measured by fall-of-potential per IEEE Std 81 on the eight nearest towers gave a mean of 27 Ω and a maximum of 42 Ω on the tower nearest the station on one line. Footing resistance dominates backflashover performance because it sets how high tower-top potential rises for a given injected current: at 42 Ω the critical flashover current is 74 kA, whose exceedance probability from the IEEE stroke current distribution is 9.4 percent; at 12 Ω the critical current is 168 kA, exceedance probability 1.2 percent. Lightning Performance of Approach Spans

The consequence of a backflashover on the approach spans is not the line outage — the line trips and recloses — but the steep-fronted surge entering the station. Incoming steepness at the station entrance was computed at 1,600 kV/µs as found, after corona damping had reduced it from about 2,800 kV/µs at the strike point over 1.5 miles of intervening line. The separation distance problem. An arrester does not clamp the voltage at the equipment it protects; it clamps the voltage at its own terminals. Between arrester and equipment lies conductor down which the surge must travel, and during that travel time the incoming wave continues to rise. The voltage at the equipment therefore exceeds the arrester discharge voltage by V(equipment) = V(arrester) + 2 · S · d / c where S is incoming steepness in kV/µs, d separation in feet and c the propagation velocity of 984 ft/µs. To this is added the inductive drop along the arrester lead and ground connection, L · di/dt, with lead inductance of 0.40 µH/ft and a representative discharge current rate of rise of 12 kA/µs, giving 4.75 kV per foot of lead. At 1,600 kV/µs the separation term is 3.25 kV per foot. One hundred fifty-one feet therefore adds 491 kV to the arrester's 412 kV discharge voltage before the lead is counted. That is the mechanism by which a nominally protected 230 kV transformer bushing sees 979 kV against a 900 kV BIL. Lightning Protective Margins, As Found

Three locations fail the 20 percent criterion and one is negative — the protective level exceeds the withstand level, meaning the equipment is not protected at all against the governing surge. That location is the bushing that failed.

7.4 Arrester duty and energy absorption

Energy capability is the quantity most often overlooked in arrester specification, because the catalog page leads with discharge voltage. Discharge voltage determines whether the arrester protects; energy capability determines whether it survives protecting. The installed 500 kV arresters are station class, duty cycle rating 396 kV, MCOV 318 kV, line discharge class 3, single-event energy capability 7.8 kJ/kV of rating or 3.09 MJ. IEEE Std C62.11-2020 has since restructured this classification, replacing line discharge classes with explicit charge transfer and thermal energy ratings against rated voltage; the study reported both formulations so replacements could be specified against either. For each of the 400 shots in the reclosing case the study integrated instantaneous arrester power over the full transient and built a distribution. The 98th percentile — matching the E2 basis used for voltage — was 4.9 MJ, or 12.4 kJ/kV, against a capability of 7.8 kJ/kV. Every reclose onto substantial trapped charge was demanding roughly 1.6 times the arrester's single-event capability. Metal-oxide arresters do not fail instantly on overduty. They degrade: absorbed energy raises block temperature, temperature raises leakage current at operating voltage, leakage raises temperature further, and if the loop does not stabilize the arrester runs away, typically minutes to hours after the causing event. Several observed failures had no coincident recorded disturbance, which had made them look random. Cross-referencing failure timestamps against the Owner's breaker operation records showed five of the seven arrester failures occurred within four hours of a reclose on the associated line. That correlation was decisive. Arrester Energy Duty

Arresters were also checked for pressure relief and fault current withstand per IEEE Std C62.11. Available fault current at the 500 kV arrester locations is 52 kA symmetrical against an installed pressure relief rating of 40 kA. That mismatch explains the violent nature of two failures and was independently a specification defect.

7.5 Insulation strength, clearances and the statistical check

Insulation levels are BIL 1800 kV and BSL 1300 kV for 550 kV maximum equipment and BIL 900 kV and BSL 745 kV for 242 kV maximum equipment, from the IEEE preferred level tables. Chopped wave withstand for the transformers is 1.10 times BIL — 1,980 kV and 990 kV — at 3 µs time to chop. The chopped wave check matters here because the governing backflashover surge is itself effectively chopped by flashover of the insulator string, and the front-of-wave arrester discharge voltage of 886 kV at 500 kV exceeds the 8/20 value. At the Bank 2 500 kV bushings the as-found chopped wave margin was 26 percent, passing, while the full-wave margin was 17 percent, failing — the two checks are not redundant. The site elevation of 3,900 feet gives a relative air density of 0.864 and requires altitude correction of self-restoring insulation. The correction exponent was taken as 1.0 for lightning impulse and 0.75 for switching impulse, giving factors of 1.157 and 1.115. Non-self-restoring internal insulation is unaffected by altitude, so transformer BIL was not corrected while air clearances and external bushing surfaces were. The governing 500 kV phase-to-ground air clearance is 13.1 feet. By the gap factor method, CFO = 500 · k · d^0.6, with d the gap in meters (4.0 m here) and a gap factor of 1.15 for the conductor-to-structure geometry present, the standard-atmosphere switching impulse CFO is 1,321 kV, reducing to 1,185 kV at site density. With a standard deviation of 5 percent of CFO, or 59 kV, V3 = CFO − 3σ is 1,008 kV. Applying the two-parameter method to the as-found reclosing case gives γ = 1,008 / 1,302 = 0.77. A statistical safety factor below unity means the 2 percent overvoltage exceeds the 0.13 percent withstand — an unambiguous failure. With controlled closing, γ = 1,008 / 853 = 1.18, above the 1.15 criterion but not by much. Because the margin was thin, the study proceeded to full risk integration, fitting the 400-shot distribution to an extreme value form and integrating against the Gaussian insulation discharge probability across the six air gaps in parallel exposure along the switched path. Statistical Risk of Insulation Failure, 500 kV Switching

The full integration gave a lower risk for controlled closing than a comparison of γ values would suggest, because controlled closing truncates the upper tail of the overvoltage distribution rather than merely shifting the whole distribution downward. Pre-insertion resistors shift; controlled closing truncates. That difference is invisible to the two-parameter method, which is why the full integration was performed. Contamination performance was checked against US practice for the measured environment. Equivalent salt deposit density sampling on removed insulators gave 0.06 mg/cm², placing the site in the light-to-moderate band and calling for approximately 20 mm/kV of specific creepage referenced to maximum line-to-line voltage. The 500 kV bushings provide 14,600 mm, or 26.5 mm/kV, and the 230 kV bushings 6,100 mm, or 25.2 mm/kV. Contamination was excluded as a contributing cause.

7.6 Surge transfer to the 13.8 kV tertiary

Surges arriving at the 500 kV or 230 kV terminals transfer to the buried delta tertiary capacitively through the inter-winding capacitance, dominant in the first microsecond, and inductively through the turns ratio thereafter. The 13.8 kV bus, its station service transformer and its cable have a BIL of 110 kV. Simulation with the measured capacitance network gave a transferred crest of 96 kV for the governing 500 kV backflashover case, a margin of 14.6 percent, below criterion. There were no tertiary arresters. Adding 13.8 kV distribution class arresters with MCOV 8.4 kV and a 10 kA discharge voltage of 28.9 kV at the tertiary bushings with leads under 3 feet reduced the transferred crest to 41 kV and raised the margin to 168 percent. This finding was incidental to the study's original purpose and among its more valuable outputs, because a tertiary failure inside an autotransformer is not a repairable event.

8. Sensitivity and Scenario Analysis

Sensitivity Analysis

Three results shaped the recommendations. Separation distance is the dominant variable in the lightning problem and also the cheapest to change: every other lightning mitigation buys tens of kilovolts of improvement, while moving an arrester sixty-five feet closer buys two hundred. The switching conclusion does not depend on an extreme trapped charge assumption — even at 0.6 pu, E2 remains 2.31 pu and arrester energy duty remains above capability, so the controlled closing recommendation cannot be attacked as worst-case reasoning. And line model choice is not a modeling nicety: Bergeron models understated E2 by 0.19 pu, enough on its own to move the as-found reclosing case from clear failure to apparent marginal pass. A study performed with constant-parameter lines would likely have concluded the station was acceptable.

9. Findings and Root Cause Assessment

Findings Register

Root cause was assigned to three independent chains, each a case of a change made competently in its own domain propagating into a domain nobody looked at. Root cause 1 — separation distance created by an undocumented bay reconfiguration. The Bank 2 230 kV arresters were originally on the bushing risers, about 15 feet from the equipment. During a bay reconfiguration they were removed to permit structure work and never reinstated; protection was thereafter provided by arresters on the dead-end structure 151 feet away. No engineering review of that substitution occurred, because to the crew and the supervising engineer the bay still had arresters on it. The drawing set was never revised. The resulting travel-time term is the direct and sufficient cause of the bushing failure, and it explains why the failure occurred on Bank 2 and not Bank 1. Root cause 2 — weakened zero-sequence source following generation retirement. Retirement removed grounded-wye-delta step-up transformers from the local zero-sequence network, raising X0/X1 at the 500 kV bus from 3.0 to 4.6 and the earth fault factor from 1.25 to 1.35. The retirement study correctly evaluated power flow, stability and short-circuit duty. It did not evaluate earth fault factor, because that is not a standard output of a retirement study and because nobody downstream was asked whether it mattered. It mattered: it consumed the entire TOV margin of arresters correctly selected under the original grounding condition. Root cause 3 — loss of pre-insertion resistors in a breaker retrofit. The replacement upgraded interrupting capability and maintainability. It also removed the pre-insertion resistors, which were not a required feature of the replacement specification because that specification was written around interrupting duty, mechanical life and gas system performance. Nothing in the procurement chain flagged that the resistors performed an insulation coordination function. The consequence is Finding F2, and F2 is the cause of five of the seven arrester failures.

10. Mitigation Options and Recommendations

Mitigation Options Comparison

Option D was recommended and adopted. Option E was rejected on the explicit basis that the statistical safety factor under D, at 1.18, already exceeds the 1.15 criterion, and that raising it to 1.29 by structural modification of the 500 kV dead-end steel returned no defensible reduction in computed risk for the incremental $2.7M. That recommendation was made in writing with the supporting risk integration, because declining to spend money is an engineering recommendation requiring the same evidentiary standard as spending it. Arrester re-specification. All twelve 500 kV and fifteen 230 kV arresters replaced. The 500 kV units retain the 396 kV duty cycle rating and 318 kV MCOV — MCOV cannot be raised without violating the continuous voltage requirement — but move to a heavy-duty design with single-event energy capability of 13.6 kJ/kV of rating (5.39 MJ), pressure relief 65 kA, 10 kA 8/20 discharge voltage 806 kV, 20 kA discharge voltage 892 kV, switching surge protective level 675 kV at 2 kA, and TOV capability with prior energy of 1.43 pu at 0.6 s and 1.42 pu at 1.0 s. The 230 kV units move from 4.5 to 7.8 kJ/kV with a 10 kA discharge voltage of 396 kV. Raising energy capability while lowering discharge voltage at the same rating is achievable with larger-diameter blocks, and is the one specification lever that improves both sides of the margin equation at once. Arrester relocation. Bank 2 230 kV arresters returned to the bushing risers at 15 feet separation with 6-foot leads. New dedicated arresters at the Bank 2 500 kV bushings at 25 feet separation with 7-foot leads, rather than continuing to rely on line entrance units 190 feet away. New arresters adjacent to the 230 kV bus instrument transformers at 39 feet. Lead routing specified for the shortest practical run with the ground lead bonded directly to the equipment base rather than to the yard grid at a distant point, so arrester and equipment share a common reference during the surge. Controlled closing, line arresters and counterpoise. Four 500 kV breakers fitted with independent pole operation controllers referenced to line-side and source-side voltage, with compensation for operating time drift with temperature and control voltage. Transmission line arresters installed on all phases of the first two towers of each incoming line; these clip the incoming surge and, equally importantly, de-steepen it, reducing computed steepness at the station entrance from 1,600 to 1,100 kV/µs and thereby improving every separation-limited margin in the yard simultaneously. Four radial counterpoise runs of 250 ft in 2/0 AWG copper-clad steel at each of the eight approach towers, designed in CDEGS against the two-layer soil model. Grounding bank, clearing time, shielding and procedures. A 230 kV grounding transformer restores the zero-sequence source, reducing X0/X1 to 3.1 and the earth fault factor to 1.27. Breaker failure backup clearing reduced from 1.1 s to 0.6 s with coordination verified. Two 80-foot masts eliminate rolling sphere penetration in the 230 kV yard. Tertiary arresters fitted. The two ferroresonance-capable switching orders withdrawn and replaced, with a VT secondary loading resistor fitted as defense in depth. Post-Mitigation Lightning Protective Margins

Every margin exceeds 40 percent; the minimum is 52.3 percent. Post-mitigation TOV duty is 1.21 pu of MCOV for 0.6 s against a capability of 1.43 pu, a ratio of 1.18 against the 1.10 criterion.

11. Implementation Support and Field Validation

Keentel supported implementation through arrester technical specification and bid evaluation, relocation general arrangement drawings and structure loading checks, the counterpoise design package, controlled closing setting calculations, and witness of commissioning tests. Implementation ran from Month 9 to Month 22, staged around bank outages already scheduled for other work. Validation was designed to test the model rather than merely confirm the installation. Field Validation Against Simulation

The controlled closing validation could have invalidated the recommendation. Forty-two operations were recorded across the four retrofitted breakers over six months, with line-side voltage captured on the station disturbance recorders at 15.4 kHz. The highest peak observed was 1.86 pu. That is a small sample against a 2 percent statistic and cannot by itself confirm the E2 value of 1.90 pu, and the report said so. What it confirms is that no operation approached the 2.9 pu of the as-found configuration, and that measured pole scatter of 0.52 ms is better than the 0.6 ms assumed, placing true E2 at or below the simulated value. The staged single line to ground test on the 230 kV bus, performed during a planned outage, measured an earth fault factor of 1.25 against a simulated 1.27 with the grounding bank in service; that test is the direct validation of the zero-sequence root cause and closed the argument.

12. Results and Value Delivered

Outcomes Scorecard

At forty-eight switching operations per year the risk figures correspond to an expected insulation failure interval of approximately 4.5 years before mitigation and approximately 870 years after. The pre-mitigation figure is consistent with the observed failure history, which is the strongest available evidence that the model was right. The capital program cost approximately $4.4M against a single-bank replacement exposure of $9.8M with a fourteen-month lead time for a full three-phase set, before any consideration of the transfer constraint a lost bank imposes.

13. Lessons Learned and Engineering Insights

Insulation coordination is a live document, not a design-stage calculation. Three separate competent engineering activities each degraded this station's insulation coordination and none had reason to know it. The remedy is not more analysis; it is a short screening checklist attached to change control asking whether a change touches arrester geometry, breaker closing behavior, zero-sequence source strength, or clearing time. Any yes triggers a review. That checklist was delivered as an unscoped addition to the report. The catalog discharge voltage is not the protective level. An arrester with an excellent 806 kV discharge voltage protects nothing at 190 feet of separation, because the separation term added 619 kV at the as-found incoming steepness. Arrester selection is a geometry problem at least as much as a product selection problem, and here relocation delivered more margin than replacement did. Removing a component is a design change even when nothing is added. The pre-insertion resistors were removed by a specification that simply did not mention them. A functional loss produced by omission is harder to detect in review than a change produced by addition, because there is no line item to review. Like-for-like replacement specifications should enumerate the functions of the removed asset, not just its ratings. Statistical methods change conclusions, not just confidence. The two-parameter method ranked pre-insertion resistors and controlled closing as near-equivalent at γ of 1.22 and 1.18. Full risk integration ranked controlled closing better by a factor of 2.5, because it truncates the distribution tail rather than shifting the distribution. Where the decision is close, the extra effort is not academic rigor; it determines the answer. Model fidelity choices are engineering decisions with consequences. Constant-parameter line models understated switching overvoltage by 0.19 pu, enough to convert a clear failure into an apparent pass. A static arrester V-I characteristic would similarly have understated lightning protective levels in exactly the marginal cases. State what the model can and cannot represent before running it, then run the sensitivity that tests the statement. Field survey beats drawing review. The most valuable hour in this study was spent with a tape measure. The as-built drawings were internally consistent, professionally produced and wrong.

14. Keentel Capability Summary

• Full insulation coordination studies to IEEE Std 1313.1 and 1313.2, deterministic and statistical, at all US transmission voltage classes through 765 kV
• Electromagnetic transient modeling in PSCAD/EMTDC and EMTP, including frequency-dependent line models, corona, nonlinear footing impedance and transformer high-frequency representations
• Statistical switching studies with randomized pole closing, distribution fitting and full risk-of-failure integration
• Lightning performance analysis: backflashover rate, shielding failure flashover rate, and station shielding to IEEE Std 998 rolling sphere and electrogeometric methods
• Surge arrester selection, specification and application to IEEE Std C62.11 and C62.22, covering MCOV, duty cycle rating, energy class, TOV capability and pressure relief coordination
• Temporary overvoltage analysis: load rejection, ground fault and coefficient of grounding, resonance, ferroresonance and transformer saturation
• Tower footing and counterpoise design in CDEGS with field measurement to IEEE Std 81
• Arrester geometry survey, separation distance and lead length analysis, and relocation engineering
• Root cause investigation of dielectric failures correlated against operating and disturbance records
• Field validation planning, staged fault testing, disturbance recorder analysis and commissioning witness
• Independent review and tool-diversity cross-checking of third-party transient studies

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.