Insulation Strength: BIL, BSL, CFO and Impulse Testing

In simple terms, insulation coordination checks whether equipment insulation is strong enough for the overvoltages expected in service. Everything else on this page — BIL, BSL, CFO and statistical withstand — is just a more precise way of answering that one question.

In high-voltage power systems, insulation must withstand different types of overvoltages. These overvoltages may come from lightning strokes, switching operations, faults and fault clearing, energisation of lines, transformers, cables or reactors, and fast-front transients in GIS or substations.

The purpose of insulation coordination is to answer one main engineering question:

To answer this, the insulation strength must be specified in a standardised way. That is why terms such as BIL, BSL, CFO, statistical withstand, conventional withstand, and the standard lightning and switching impulses exist. The sections below build these up in the order they are best understood — insulation types first, then the withstand levels, then how they are proven and corrected for real conditions.

Insulation is classified in two independent ways: by its physical location (external or internal), and by what happens after a flashover (self-restoring or non-self-restoring). Both classifications matter, because together they decide how the insulation can be tested.

An air gap at sea level has a higher insulation strength than the same gap at high altitude, because air density is lower at altitude. So external insulation is not a fixed number — it depends on environmental conditions.

BIL is the insulation strength against a standard lightning impulse (in IEC language, the lightning impulse withstand voltage). Key points: it is a crest value in kV, tied to a standard lightning impulse waveshape, defined under standard atmospheric conditions, and generally associated with dry conditions.

The standard lightning impulse waveform is \(1.2/50\ \mu\text{s}\): front time ≈ 1.2 µs (how fast the voltage rises) and time to half-value ≈ 50 µs (how long the tail lasts before decaying to half the crest).

BSL is the insulation strength against a standard switching impulse (the switching impulse withstand voltage). The standard switching impulse waveform is \(250/2500\ \mu\text{s}\): front time ≈ 250 µs, time to half-value ≈ 2500 µs. Switching impulses are much slower than lightning impulses.

The simple memory rule: BIL = lightning stress, BSL = switching stress.

The same words — “front” and “tail” — are used for both impulses, but the front time is measured differently.

Lightning impulse \((1.2/50\ \mu\text{s})\)

The voltage rises very quickly to the peak and then decays. The time to crest is not measured from the real zero, but from a virtual origin: a straight line is drawn through the 30% and 90% points of the front, and where it crosses the time axis is the virtual origin. The front time is measured from there to the crest. This avoids the problem that real waveforms are not perfectly smooth at the very beginning.

Switching impulse \((250/2500\ \mu\text{s})\)

The front time is measured from actual zero to actual crest — not from a virtual origin. So although the terminology is the same, the measurement method differs from the lightning impulse.

CFO — Critical Flashover Voltage is the voltage at which self-restoring insulation has a 50% probability of flashover. If an impulse with crest equal to CFO is applied many times, about half the shots flash over and half withstand. CFO is used mainly for self-restoring insulation such as air gaps and external insulation.

For self-restoring insulation the flashover probability is often assumed Gaussian. The CFO is the mean of the distribution, and the BIL or BSL sits at the 10% flashover point, giving:

In plain terms, this means the specified withstand level is intentionally placed below the average flashover voltage, so that only a limited probability of flashover (about 10%) is expected for self-restoring insulation at the rated level.

The spread is often expressed as a coefficient of variation \(\sigma/\text{CFO}\). Typical values:

Example. An air gap with statistical BIL of 1050 kV: at a 1050 kV standard lightning impulse, about 90% of shots withstand and about 10% may flash over — acceptable, because the insulation recovers.

One caveat: the Gaussian assumption extends infinitely to the left, implying a tiny flashover probability even at very low voltage, which is not physically true. It is nevertheless accepted down to a few standard deviations below CFO; the Weibull distribution is more realistic, but most historical insulation data use Gaussian assumptions.

For self-restoring insulation, tests may allow a limited number of flashovers. Three common procedures, all applied at a crest equal to the specified BIL or BSL:

No real test perfectly separates good from bad insulation. Suppose the actual flashover probability is 20% (worse than the intended 10%): statistical uncertainty means the equipment may still pass — but the chance of that depends on the method. The 3+3 test gives the highest chance of bad insulation passing; 2/15 the lowest. From an insulation-quality viewpoint:

Standards provide preferred values so equipment ratings are standardised, but the exact value depends on system voltage, equipment type, insulation type, the standard followed, manufacturer practice and utility requirements. As examples:

• Transformers, bushings and GIS have defined BIL (and BSL) levels.
• Cables usually have BIL but not necessarily BSL.
• Circuit breakers may have different BSL for the open and closed positions.

For transformers, BSL is often about 83% of BIL, though exact values depend on the standard and equipment class.

For example, when a selected equipment withstand level is compared with the representative overvoltage obtained from EMTP® / EMT studies, the insulation coordination margin can be checked — this is the step that links these preferred standard values back to a real power-system study.

To find CFO, impulses are applied at several crest voltages. At each level a number of shots is applied, the flashovers are counted, and the percentage flashover is computed — for example, 2 of 100 shots at 900 kV (2%), 20 of 40 at 1000 kV (50%), rising further at higher voltage. The results are plotted on Gaussian probability paper; the voltage at 50% flashover is the CFO, and σ comes from the difference between the 50% point and the 16% or 84% point.

Once CFO and σ are known the strength curve is fixed. For CFO = 1000 kV and σ = 50 kV, the coefficient of variation is \(\sigma/\text{CFO}=5\%\), and the 10% flashover (statistical withstand) level is:

This 936 kV is the statistical BIL or BSL (depending on the impulse type).

A chopped wave is a lightning impulse intentionally collapsed after a short time — the voltage rises like a lightning impulse, then is suddenly forced down, usually by flashover of an external chopping gap. The rapid collapse produces a very severe, steep stress, which is especially important for transformer turn-to-turn insulation because it creates steep internal voltage gradients in the windings.

The key point is that the severity is not only due to the peak voltage, but also due to the very rapid voltage collapse. This sudden change produces steep voltage gradients inside transformer windings, stressing the turn-to-turn insulation far more than a smooth full-wave impulse of the same crest would.

Chopped-wave tests are mainly specified in ANSI standards, not IEC. Originally they represented a surge entering a substation being suddenly chopped by flashover of nearby insulation. That exact scenario is not always considered realistic today, but the test remains valuable: it stresses insulation under very fast voltage changes — transformer turn-to-turn insulation in particular, and it is relevant for GIS-connected transformers where disconnector operation can generate very fast front transients.

Time-lag (volt-time) curves

A time-lag (volt-time) curve plots applied crest voltage against time to flashover: apply increasing levels, record when flashover occurs, and plot voltage versus time. For self-restoring insulation, higher voltage usually produces faster flashover, and the curve flattens at longer times — for air, the asymptote corresponds approximately to the CFO. The point is that insulation strength depends on the duration of the applied voltage: short impulses need higher voltage to flash over, longer ones flash over at lower voltage.

BIL, BSL and CFO are defined at standard atmospheric conditions, but real substations and laboratories are not always at those conditions. Because external insulation strength changes with air density and humidity, an atmospheric correction is needed. The factors that matter are relative air density, humidity, altitude, temperature, pressure, and whether conditions are wet or dry.

Relative air density depends on pressure, temperature and altitude. At high altitude, pressure and therefore air density are lower, so air-gap and external insulation strength is lower. A practical approximation:

At sea level \((A=0)\), \(\delta\approx 1\); at higher altitude \(\delta<1\), so external insulation strength decreases exponentially with altitude.

Humidity affects impulse flashover strength. For dry impulse testing a humidity correction may apply; for wet or rain conditions the humidity correction factor is \(H_c = 1.0\). Switching impulse design and testing are often based on wet conditions, and outdoor lines and substations are normally assessed wet.

For lightning impulse the correction is simpler: with wet design \(H_c=1\) and the strength mainly reduces with air density — so at higher altitude the same clearance has lower lightning-impulse strength. To maintain the withstand capability you must increase the clearance, raise the standard insulation level, or specify altitude correction.

For switching impulse the correction can be more complex because the exponent \(m\) depends on a parameter \(G_o\) (related to strike distance and the standard-condition CFO). The workflow is: assume wet conditions, determine the required strength at site altitude, convert it to an equivalent standard-condition value, and select the standard BSL accordingly — sometimes by iteration.

Iterative example. To convert a required switching-impulse CFO at site altitude into an equivalent standard CFO: assume a standard CFO → compute \(G_o\) → determine \(m\) → compute the correction factor → update the standard CFO → repeat until it converges. This is a common insulation-coordination workflow when site conditions are non-standard.

Lightning and switching impulses are produced by impulse generators, most commonly the Marx generator, which charges capacitors in parallel and discharges them in series. In sequence: the capacitors charge through resistors; a trigger gap fires; the first spark gap breaks down; the voltage across the next gap rises; sparkover cascades through all stages; the capacitors are effectively connected in series; and a high impulse voltage results. With many stages, the output can be very high.

The waveform is controlled by the generator capacitance, series resistance, load capacitance, discharge resistance, divider circuit and stray inductance. The front is mainly set by the series resistance and capacitance; the tail by the discharge path and divider resistance. A switching impulse needs a much longer front and tail, so larger external shaping resistors are used. An impulse voltage is often represented as a double-exponential waveform:

Generator efficiency is the crest output voltage divided by the open-circuit charging voltage. Longer wavefronts need larger series resistance, which reduces the output crest — so producing switching impulses may need more generator capability or different shaping arrangements.

High impulse voltages cannot be measured directly, so a voltage divider reduces them to a recordable value: resistive dividers for lightning impulses, capacitive dividers for switching impulses, and RC dividers for mixed/broader use.

The five core terms at a glance — including what each abbreviation actually stands for:

For practical insulation-coordination studies, do not simply compare a simulated overvoltage with a nameplate level. The correct process: