A factory acceptance test is the last good chance to catch a problem before a transformer becomes the utility’s problem. Once it leaves the factory, a defect that could have been a warranty conversation in the test bay becomes a field failure, an outage, and a multi-month replacement. The witness in the chair is the line between those two outcomes.
Witnessing a FAT well is a different skill from understanding the standards. The standards tell you what the tests are and what the limits are. Witnessing is about reading the room, the records, and the manufacturer’s behavior in real time — knowing which results are solid, which are marginal, which are being quietly explained away, and when to stop the test and ask a question.
This is a walkthrough of a power transformer FAT from the witness chair. What happens, in roughly the order it happens, and what matters at each stage. It assumes familiarity with the dielectric tests themselves — the focus here is on the witnessing, not the theory.
Table of Contents
Before Anything Is Energized
The most useful work a witness does happens before the first test. By the time voltage is applied, most of the leverage is gone.
The test plan review. The manufacturer issues a test plan — the sequence of tests, the levels, the acceptance criteria, the instruments. This should match the purchase specification and the applicable standard. Check that every test the spec requires is in the plan, at the right level, on the right windings. Tests get dropped from plans, sometimes by oversight, sometimes because the manufacturer reads the spec differently. The plan review is where a missing IVPD or a wrong impulse level gets caught while it’s still cheap to fix.
Pay attention to the test sequence. IVPD must be last among the dielectric tests — its sensitivity depends on a settled thermal and dielectric state, and tests run after it could mask or trigger PD that the test should have caught. If the plan has anything substantive between IVPD and shipment, ask why.
Calibration certificates. Every measuring instrument in the dielectric tests should have current calibration. The impulse measuring system, the PD detector and its coupling capacitor, the voltage dividers. Out-of-cal instruments invalidate results. This is a paperwork check, but it’s the foundation everything else stands on — a beautiful PD result on an uncalibrated detector means nothing.
The transformer’s condition. The unit should be fully assembled as it will operate, filled with the same oil specification it will ship with, and processed (dried, degassed, impregnated). Ask when oil filling and processing finished. A transformer tested too soon after filling hasn’t fully impregnated — the oil hasn’t penetrated the paper, and the dielectric tests, especially PD, can read worse than the finished unit will, or can mask defects. Conversely, a unit rushed through processing to hit a test date is a unit worth watching closely.
Oil samples for DGA. Standard practice on larger transformers is to pull an oil sample for dissolved gas analysis before the dielectric tests and again after. This isn’t required by IEC 60076-3, but a witness should want it. The before-and-after comparison is the single best evidence of whether the dielectric tests caused any internal distress. Gases that weren’t there before and are there after — particularly acetylene, the signature of arcing — tell you something broke down inside even if the oscillograms looked clean. Confirm the samples are being taken and that you’ll see the results.
The Routine Electrical Tests
Most FATs open with the lower-drama routine measurements before moving to the dielectric tests. These are quick, but they’re not throwaway — they establish the transformer’s baseline and catch gross manufacturing errors.
Winding resistance. Measured on every tap. Watch for balance between phases — a significant imbalance on the same tap suggests a connection problem, a brazing defect, or a winding error. The absolute values matter for loss calculations later; the phase-to-phase comparison matters for catching defects now.
Turns ratio (voltage ratio). Measured on every tap, every phase. The measured ratio should match the calculated ratio within tolerance (typically 0.5%). A ratio error means a winding turns count is wrong or a tap connection is misplaced. This is a fundamental error — if the ratio is wrong, the transformer is built wrong. Watch every tap, not just the principal one; tap-changer wiring errors show up on specific taps.
Magnetizing current and no-load loss. The excitation test. Watch the magnetizing current pattern across phases — the outer phases typically draw more than the center phase on a three-limb core, so a symmetric pattern can itself be a flag. A sudden jump in no-load loss versus the design value or versus sister units points to core problems: shorted laminations, joint issues, assembly damage.
Load loss and impedance. The short-circuit test. Impedance should match the nameplate within tolerance — it affects how the transformer shares load and behaves under fault. Load loss feeds the capitalized-loss penalties in many contracts, so the number has commercial weight. Watch that the test is done at the right tap (usually principal) and corrected to reference temperature correctly.
These tests rarely produce drama, but a problem here is usually fundamental. A ratio error or a no-load loss anomaly is a reason to pause the whole FAT and understand the cause before applying high voltage to a unit that may be built wrong.
The Applied Voltage Test
The first of the dielectric tests, and the simplest to witness. A single power-frequency voltage applied to all terminals of a winding at once, against everything else grounded, for 60 seconds.
What to watch: the voltage reaches the specified level, holds for the full minute, and comes down without any collapse. A collapse — voltage suddenly dropping as the insulation breaks down to ground — is an unambiguous fail. There’s little subtlety here; the test either holds or it doesn’t.
The thing to confirm is the level and the duration. The test value is the specified withstand voltage, and the standard defines it via the peak measurement divided by √2. Watch that the full minute is timed from when full voltage is reached, not from when the operator started ramping. A short hold isn’t a valid test.
For a non-uniformly insulated winding, remember the AV test only stresses to the neutral insulation level — the line-end insulation gets verified later by the induced or switching test. A clean AV result on a graded winding doesn’t mean the line end is proven.
The Induced Voltage Withstand Test
Now the transformer is energized through its own magnetic circuit at higher-than-rated frequency. The witnessing is similar to AV in spirit — confirm the level, confirm the duration, watch for collapse — but a few things deserve attention.
The frequency and duration relationship. Because the test runs above rated frequency to avoid core saturation, the duration shortens at higher frequencies per the standard’s formula. Confirm the duration matches the frequency used. An operator running at a high frequency for the long-duration time is over-testing; running too short is under-testing.
The voltage on the non-tested terminals. In a three-phase induced test, symmetrical voltages appear at all line terminals. Confirm the arrangement produces the intended stresses, particularly phase-to-phase, which the AV test never touches.
The IVW test is a withstand test — pass/fail on holding the voltage. It’s often combined with the IVPD test, which is where the real witnessing happens.
The IVPD Test: The One That Earns Its Hour
This is the most demanding test to witness and the one where a witness’s attention matters most. It runs over an hour, the result is a continuous trace rather than a single number, and the acceptance criteria are about pattern and trend, not just a ceiling.
The background check first. Before the transformer is at any meaningful voltage, the PD detector reads the ambient. The standard requires background PD not to exceed 50 pC for the test to be valid (100 pC for shunt reactors). Watch this. A test bay near welding, switching power supplies, or radio sources can have a background that swamps the measurement. If the background is high, the test is not valid no matter what the transformer does — and a manufacturer eager to proceed may want to start anyway. A high background is a reason to wait for cleaner conditions, not to proceed and hope.
The step-up. The voltage climbs through the measurement points — 1.2×Ur/√3, the one-hour measurement level, then the enhancement level. Watch the PD inception. At what voltage does PD activity start? A transformer that’s quiet up to the enhancement level and then shows brief activity is in a different situation from one that’s discharging steadily from just above operating voltage. Note the inception and extinction voltages — the standard expects them recorded, and they tell you about the nature of any defect.
The enhancement hold. Sixty seconds (300 for the largest units) at the highest voltage. This is the stress that’s meant to provoke any latent defect. It’s not a measurement point — it’s the trigger. Watch what the PD does during this hold and, critically, what it does immediately after when the voltage drops back to the one-hour level.
The hour. This is where witnessing becomes endurance. The PD level is measured every five minutes at the one-hour voltage. What matters is the trend. The acceptance criteria are: no level above 250 pC, no rising trend, no sudden sustained increase in the last 20 minutes, no increase greater than 50 pC across the hour, and not more than 100 pC at 1.2×Ur/√3 after the hour.
A steady trace at, say, 120 pC for the full hour is a clean pass. A trace that starts at 80 pC and is at 200 pC by minute 50 is a fail on the rising-trend and 50-pC criteria even though it never touched 250. This is the trap: a witness watching only for the 250 pC ceiling misses the trend failure. The rising trend is often the more meaningful result — it’s the signature of a defect that’s actively progressing.
Watch for the conversation that starts when a trace creeps upward. “It’ll settle.” “That’s just the bay.” “We can extend the hour and it’ll come good.” The standard does allow extending the hour — criteria (c) and (d) are met if satisfied over any continuous one-hour period — so a genuine settling is legitimate. But a trace being talked upward into acceptability deserves skepticism. The numbers are the numbers.
The post-hour check. After the hour, the voltage drops to 1.2×Ur/√3 and PD is measured again. The 100 pC limit here is the recovery check — after all that stress, the transformer should be quiet at near-operating voltage. A unit that passed the hour but reads high here has been changed by the test, which is arguably worse than a unit that struggled during the hour and recovered.
The Impulse Tests
Lightning impulse (and chopped wave, where specified) is fast, dramatic, and the witnessing is all about the records. The generator fires, there’s a sharp crack, and the result is on the recorder.
The reference impulse matters most. The sequence is one reduced-level reference impulse followed by full-level impulses. The reference is the fingerprint — the transformer’s response when (presumably) nothing has broken down. Every full-level impulse is judged against it. Watch that the reference is taken cleanly and recorded properly, because everything downstream is a comparison to it.
The comparison is the test. The acceptance is whether the full-level records match the reference in shape, allowing only for the difference the voltage change explains. Watch the neutral current trace especially — it’s the sensitive channel. Healthy: the traces overlay, scaled. Faulted: the full-level trace shows high-frequency pulse trains or a waveshape change the reference doesn’t have.
This is where the witnessing gets technical and where manufacturers and witnesses sometimes disagree. Small differences between reference and full traces are normal and explainable. The judgment is whether a given difference is benign or a sign of internal breakdown. A manufacturer’s impulse specialist will have a view; a good witness has enough independent understanding to evaluate it rather than accept it. When a difference is genuinely ambiguous, additional impulses and a careful look at the time-expanded traces are warranted before anyone declares a pass.
Watch for the re-shoot. If an impulse produces an ambiguous or concerning trace, the question of repeating it comes up. A repeat at the same level that produces a clean, repeatable trace can be legitimate evidence the first was an artifact. But repeated impulses at reducing margins, or selective presentation of the “good” shots, is gaming. A witness should track every impulse fired, not just the ones offered as the record. (This judgment call is involved enough to deserve its own treatment — the short version is: count the shots, and be wary of a fault explained away by a do-over.)
The chopped wave. Where specified, the chopped impulses are bracketed by full waves so that damage from the chop shows up in a before-and-after full-wave comparison. Watch that the bracketing full waves are actually compared, not just recorded. The chop itself is on a fast time base — confirm the records have the resolution to evaluate the front behavior.
The Switching Impulse Test
Slower than lightning, applied to provoke the insulation-to-earth and along-the-winding stresses that switching surges produce. The witnessing peculiarity here is the core.
Because the switching impulse is long enough for flux to build, the core saturates partway through, limiting how long the voltage can be sustained. Manufacturers handle this with reverse-polarity priming impulses before each test impulse, resetting the core magnetization so the full impulse can develop. Watch that the priming is being done and that the resulting impulse meets the required time-to-zero. Successive switching impulse oscillograms can legitimately differ because of saturation effects — unlike lightning impulse, where the traces should overlay. This means the switching impulse comparison is read differently, and a witness should understand that a difference here isn’t automatically a fault the way it would be on a lightning trace.
Temperature Rise, If In Scope
The heat run is long — many hours — and often witnessed only at setup and key checkpoints rather than continuously. If it’s in the scope, confirm the test method, the loading, and that the thermocouples and measurement points are where they should be. The result feeds the transformer’s rating and its life expectancy. The witnessing here is mostly about confirming the setup is honest and the final temperatures are read correctly; the endurance of sitting through the run is usually delegated.
After the Tests: Closing the Loop
The second DGA sample. Pull the after-test oil sample and compare to the before. This is the quiet check that catches what oscillograms miss. Acetylene appearing post-test is evidence of arcing somewhere during the dielectric tests, even if every individual test was declared a pass. Hydrogen and hydrocarbon increases point to other distress. A clean before-and-after DGA is meaningful corroboration that the unit came through the high-voltage tests without internal damage.
The records package. Every test record, every oscillogram, every PD trace, the calibration certificates, the DGA results. The witness’s job includes confirming the records are complete and consistent with what was actually observed in the bay. A record that doesn’t match what happened — a trace that’s cleaner than what was on the screen, a duration that’s longer than what was timed — is a problem worth raising before sign-off, not after.
The sign-off. Witnessing ends with accepting or not accepting the results. The leverage that existed before energization is gone, but the decision to accept is still real. A unit with a marginal PD trend, an ambiguous impulse trace, or an unexplained DGA shift is a unit to question before signing, while the manufacturer still owns it and the cost of resolution is theirs.
The Pattern Across the Whole FAT
The tests that produce a clean number — ratio, resistance, AV withstand — are easy to witness and hard to fudge. The tests that produce a trace or a trend — IVPD, impulse — are where witnessing skill matters, because the judgment is in reading the record and the room, not just checking a value against a limit.
The recurring failure mode for a witness isn’t missing a dramatic breakdown — those announce themselves. It’s accepting a marginal result that’s been smoothed over: a PD trend talked into acceptability, an impulse fault explained as an artifact, a sequence shortcut justified by schedule, a borderline result presented as comfortably passing. The defense against all of these is the same — know what the test is actually proving, watch what actually happens rather than what’s reported, and treat “it’ll be fine” as a reason to look harder, not a reason to relax.
A transformer that genuinely passes a properly witnessed FAT is a transformer with a strong claim to forty years of service. A transformer that passed because nobody looked too hard at the marginal results is a field failure waiting for its schedule. The difference is the witness.