Power Transformer Testing

A transformer is not approved because it passed. It is approved because nothing moved.

The power transformer is the richest equipment family in all of high voltage testing, and for exactly that reason it is the one that best teaches the routine, type and special logic built in Module 11. No other test object combines, in a single factory flow, precision metrology (turns ratio to ±0.5%), power measurement at extremely low power factor (losses), dielectric tests of four distinct natures (applied, induced, lightning impulse and switching impulse), a thermal test that runs for days, and a set of mechanical, chemical and acoustic verifications.

A transformer leaving the factory carries a promise of 30 to 50 years of service. Factory testing is the only moment when that promise can be verified under controlled conditions, with the object still accessible, instrumented and under the manufacturer's contractual responsibility. After energization every check becomes more expensive, riskier and less conclusive. So the factory sequence is not a task list. It is a courtroom, each test is a witness, and the order of testimony matters. The fundamental pattern is the reference sandwich: reference measurements before the tests that stress the insulation, dielectrics in the middle, the same reference measurements repeated afterwards. The numerical comparison between before and after is the verdict.

What you will be able to do

• Describe the complete factory routine test sequence for a power transformer and justify the position of each test in the order of execution, recognizing the sequence as an evidence-preservation algorithm.

• Perform and interpret the reference measurements — turns ratio, polarity and phase displacement, winding resistance, insulation resistance, capacitance and tan delta — applying the tolerances of IEC 60076-1:2011 and the relevant temperature corrections.

• Measure no-load and load losses with the normative corrections, relating each loss component to its physical origin and to the contractual guarantees.

• Distinguish conceptually and operationally between the applied voltage test and the induced voltage test, explaining which insulation each one stresses and why the induced test requires elevated frequency.

• Parameterize an induced voltage test with partial discharge measurement, including the stepped voltage profile, the duration as a function of frequency, and the acceptance criteria in pC.

• Interpret full and chopped lightning impulse oscillograms by the reduced-wave comparison method, identifying inter-turn failure signatures.

• Describe the type test (temperature rise) and special tests (switching impulse, short circuit, sound level, frequency response, dissolved gas analysis) with their quantitative acceptance criteria.

• Audit a factory routine test report, checking sequence, criteria and internal consistency of results.

The sequence is an algorithm, and it is a chain of custody

Reference measurements, then losses, then dielectrics, then the reference measurements again. Every departure from that order invalidates comparisons. Losses measured after the impulse carry the uncertainty of whatever the impulse may have done. A turns ratio measured only at the end does not prove the ratio was the same before the dielectric tests. A report that scrambles the order is not merely disorganized, it is technically weaker, because it breaks the chain of custody of the evidence. A transformer is not approved because it passed the dielectric tests; it is approved because it passed them and the reference measurements did not move.

The taxonomy of Module 11 applies here with maximum precision, and the boundary between categories shifts with voltage class. Under IEC 60076-3:2013 the lightning impulse is a routine test for Um above 72.5 kV and the switching impulse is a routine test for Um above 170 kV, which reinforces the earlier lesson: the category is not a property of the test, it is a property of the test-and-equipment pair. The level tables are coordinated per Um. At 24 kV: 125 kV full lightning impulse, 138 kV chopped, 50 kV applied. At 145 kV: 550 kV, 605 kV and 230 kV applied. At 245 kV: 950 kV, 1,045 kV, 750 kV switching impulse and 395 kV applied. At 550 kV: 1,550 kV, 1,705 kV, 1,175 kV switching and 680 kV applied. The chopped wave crest is 110% of the full impulse throughout.

Reference measurements: ratio, resistance and the demagnetization rule

Turns ratio is the first witness. It verifies that the transformer was wound and connected the way the design says, measured on every tap position, phase by phase, by a ratio bridge that applies reduced voltage to one winding and reads the induced voltage in the other. The IEC 60076-1:2011 tolerance for no-load ratio on the principal tap of the first winding pair is the smaller of ±0.5% of the declared ratio and ±10% of the actual percentage impedance, and in practice ±0.5% governs almost always. Polarity and vector group are verified at the same time, since a modern three-phase instrument reads phase displacement directly in degrees and a wrong group shows up as 30° or 60° off. The classic errors: confusing turns ratio with rated voltage ratio in star-delta connections and losing the √3, measuring with the tap changer on a different position than recorded, reading instability caused by residual magnetism from an earlier DC measurement, and forgetting that the ratio moves typically 1.25% per step on a tap changer. A real deviation on one isolated tap points to an internal tap connection error; a deviation on every tap points to turns added or missing, which is a return-to-manufacturing defect.

DC winding resistance on every winding and every tap anchors three things: the separation of load losses into I²R and stray components, the temperature rise test by resistance variation, and the screening of loose connections. Phase-to-phase agreement should sit within 1 to 2%, and one phase consistently higher points to a loose connection at the tap changer, the bushing or an internal joint. Two execution cautions dominate. Settling: the winding is an inductor with an L/R time constant reaching tens of seconds to minutes on large cores, and the reading is valid only when the current is fully settled, with a practical criterion of less than 0.1% variation per minute. Instruments that inject simultaneously into HV and LV cut that time dramatically by saturating the core. Temperature correction: R2 = R1·(k + θ2)/(k + θ1) with k = 235 for copper and 225 for aluminium, referred to 75 °C for losses, and valid only if the winding was in thermal equilibrium — three hours of rest without excitation is factory practice. And the golden rule of the flow: demagnetize after any DC measurement. Direct current leaves remanence that distorts excitation current, generates anomalous inrush and shifts impulse oscillograms.

Losses: two measurements, two guarantees

With one winding energized at rated voltage and frequency and the others open, no-load loss is core loss (hysteresis plus eddy current in the grain-oriented steel) plus minor dielectric and stray terms. The measurement uses a precision wattmeter with reference instrument transformers, and the voltage must be read with a mean-value voltmeter scaled by 1.111 in parallel with an RMS voltmeter: the ratio between the two readings reveals source waveform distortion, and IEC 60076-1 gives the correction when the distortion is moderate. A weak source, saturated by the transformer's own non-sinusoidal excitation current, measures the wrong losses. Excitation current in modern units is remarkably low, 0.2 to 2% of rated, and extremely sensitive to magnetization history: a core with remanence shows asymmetric excitation current and inflated apparent losses that relax over minutes. Run the test after demagnetization and repeat it after the dielectric tests, because shorted turns raise no-load loss unmistakably. Tolerances: +15% on the declared no-load loss component provided total losses do not exceed +10%, and +30% on declared excitation current.

With one winding solidly short-circuited and the other carrying rated current, or a fraction of it with quadratic correction, you measure short-circuit impedance and load loss for each winding pair on the principal and extreme taps. The power factor is extremely low, cos φ between 0.01 and 0.05 on large units, which puts the metrology chain of Module 10 at its limit and demands wattmeters with certified angular error. Load loss separates into I²R, computed from the resistances above, and stray loss from eddy currents in conductors and structural parts. The correction to 75 °C runs in two opposite directions, since I²R rises with temperature while stray loss falls. Short-circuit impedance on the principal tap carries a tolerance of ±7.5% of declared when Z ≥ 10% and ±10% when Z < 10%, and it is a hard contractual guarantee because it governs system fault current and load sharing in parallel. The measured reactance is also the thread of mechanical diagnosis later: it is the reference for the post-short-circuit deformation criterion and the numerical complement to the frequency response signature.

Applied is not induced, and the difference is the whole point

The applied voltage test stresses the main insulation, phase to ground and between windings. Every terminal of the winding under test is shorted together and connected to the source; all other windings, the core and the tank are earthed. Power-frequency sinusoidal voltage is applied for 60 s at the level in the insulation table for that winding's Um. Because all terminals sit at the same potential, there is no turn-to-turn stress at all: the test proves only the insulation to ground — clearances to tank and core, between windings, oil ducts and pressboard barriers. For neutrals with graded insulation intended for solid earthing, the applied test is limited to the neutral level, and verification of the line end migrates entirely to the induced test. That is the technical reason both tests coexist. The acceptance criterion is binary, no breakdown or voltage collapse, but an experienced operator monitors leakage current and acoustic noise through the minute, because audible crackle or rising current is the warning that justifies stopping before destructive failure.

The induced test proves everything the applied test cannot see: insulation between turns, between layers, between coils and between windings, with the real voltage distribution that only induction produces. Excite from the low-voltage side above rated voltage and every point of the HV winding takes the potential its turn count assigns it, so the turn-to-turn gradient is real and proportional to the enhancement. Raising induced voltage to 1.8 times rated at mains frequency is impossible, because core flux would rise in the same proportion and the core saturates long before — design flux density sits at 1.6 to 1.75 T while grain-oriented steel saturates near 2.0 T. The solution is to raise the frequency: at 100 to 200 Hz the same voltage needs half or a quarter of the flux, since U = 4.44·f·N·Φ. The normative counterpart is that when f exceeds twice the rated frequency, the enhancement duration shortens to t = 120 × fn/f seconds and never below 15 s. On a 60 Hz system tested at 200 Hz, that is 36 s.

IVPD: the verdict on workmanship

IEC 60076-3:2013 reorganized the inherited naming into two tests: the induced voltage withstand, routine on all units, with the enhancement shortened by frequency and no PD requirement; and the induced voltage test with partial discharge measurement, routine for Um above 72.5 kV. The IVPD profile is a choreography in five acts. Rise to 1.2·Ur/√3 and measure PD for at least one minute. Raise to the enhancement U1 = 1.8·Ur/√3 for the 120 × fn/f duration, minimum 15 s. Drop to the measurement level of 1.58·Ur/√3 and hold it for 60 minutes with continuous PD recording on every capacitive terminal. Return to 1.2·Ur/√3 and measure PD again. Validity requires background noise at or below 50 pC at the start and at the end, and the measurement follows IEC 60270 with calibration in pC injected at the terminals. Rise while already measuring, so you capture inception and extinction voltages.

Acceptance: no voltage collapse; during the hour, PD not exceeding 250 pC, with no rising trend, no increase greater than 50 pC across the hold and no sustained jump in the last 20 minutes; and in the final 1.2·Ur/√3 window, PD at or below 100 pC. Good factory practice works to tighter internal margins. The value of the test is in what it detects: a forgotten metallic particle, an air bubble from incomplete impregnation, a badly finished solder crest, displaced pressboard. None of those appear in the applied test, because the mean field is too low in the critical regions, and none of them collapse the induced test, because they are not disruptive yet. They appear as picocoulombs on the measurement plateau. IVPD is therefore, in practice, the quality examination of assembly workmanship, and rising PD at minute 40 is the classic signature of a migrating bubble or localized moisture. At Um of 245 kV and above, the difference from the withstand test is not the level but the temporal statistics: 60 minutes gives marginal defects time to express themselves as a trend, and the analysis requires the pC record across the whole hour, not a spot reading.

Impulse, heat, and the witnesses that stay in the oil

The lightning impulse applies the 1.2/50 µs wave to each line terminal with the others earthed directly or through low-value resistors. The sequence is one reduced full wave at 50 to 70% of the level, then three full waves at 100%; when the chopped wave is specified, it expands to reduced, one full at 100%, one or two chopped at 110% with the chop between 2 and 6 µs, then two full at 100%. Evaluation is by comparison, not by a number: overlay the scaled reduced-wave record on the full-wave records for terminal voltage and neutral current. The winding is an LC network and its impulse response is a unique signature. Any divergence — a new oscillation in the tail of the neutral current, a partial voltage collapse, a shift in oscillation frequency — reveals a change in that internal network. Inter-turn failure is the most insidious case, because it may not pull the terminal voltage down at all and only displaces the neutral current, which is why neutral current is the sensitive channel of the comparison method. Distinguishing an internal failure from an external protective gap firing is the fine art: the external flashover collapses the voltage instantly to zero with violent current, while an inter-turn failure deforms without collapsing. The switching impulse, routine above Um = 170 kV, uses a wave with crest time of at least 100 µs, at least 200 µs above 90% and at least 1,000 µs to zero, applied in negative polarity, and it explores the near-uniform stress distribution along the winding that complements the front-dominated distribution of the lightning impulse.

The temperature rise type test uses the total-loss short-circuit method: with one winding shorted, inject current so the absorbed power equals the sum of guaranteed no-load and load losses, and hold until the top-oil rise varies by less than 1 K per hour sustained over three hours. Then reduce to rated current for one hour and shut down, deriving mean winding temperatures by resistance variation with readings extrapolated back to the instant of switch-off. IEC 60076-2:2011 limits for oil-immersed units with a preservation system: top-oil rise at or below 60 K, mean winding rise at or below 65 K, hot spot at or below 78 K. Valuable by-products are the winding-to-oil gradient, the thermal time constant, and the real performance of the cooling system. A gradient larger than design points to blocked oil ducts or blocking washers left in place during assembly, a defect no electrical test sees.

Three diagnostic tests close the file. Short-circuit withstand, when contracted, applies full asymmetric current and is judged geometrically but measured electrically: the change in short-circuit reactance before and after must not exceed 1% for large circular concentric windings, 2% for circular concentric windings generally, and 4 to 7.5% for non-circular and layer windings of lower categories. Reactance is the geometry sensor, because turns that moved change the leakage inductance. Sweep frequency response measures the transfer function of each winding from 20 Hz to 2 MHz and archives it as a factory signature: below about 2 kHz the magnetizing inductance dominates and is sensitive to core condition and shorted turns, 2 to 20 kHz shows inter-winding coupling and global geometry, and 20 kHz to 1 MHz shows series capacitance and localized axial or radial displacement. The signature is only worth having if the setup is reproducible, with cables, shield earthing and tap position all recorded. And dissolved gas analysis closes the courtroom: samples before and after the dielectric campaign answer the question no oscillogram answers alone. The criterion is the increment, not the absolute. Acetylene appearing after the dielectric tests is evidence of arcing somewhere, even if every individual test was marked as passed. Hydrogen with methane points to internal discharge; ethylene with ethane points to localized overheating. New units should ship with combustible gases near zero, and any measurable acetylene increment warrants investigation before shipment. Dissolved gas analysis is less sensitive than PD measurement for incipient discharge, but it is the only witness that stays inside the object.

Syllabus

• The factory flow as an evidence-preservation algorithm and the reference sandwich

• How the routine, type and special boundary moves with voltage class

• Insulation resistance and polarization index on a large oil-paper mass

• Capacitance and tan delta of windings and bushings as the lifetime baseline

• Tightness and accessory function checks before the dielectric tests

• Turns ratio, polarity and vector group: the ±0.5% tolerance and the classic reading errors

• Winding resistance: current limits, L/R settling, demagnetization discipline, temperature correction

• No-load loss and excitation current: mean-value voltmeter, waveform distortion, remanence sensitivity

• Load loss and short-circuit impedance: low power factor measurement and the two-directional temperature correction

• Applied voltage: what it stresses, what it cannot stress, and graded insulation neutrals

• Dielectric test level tables by Um: full and chopped lightning impulse, switching impulse, applied voltage

• Induced voltage: why elevated frequency, and the t = 120 × fn/f duration rule

• The IVPD profile in five acts and the 250 pC, 50 pC and 100 pC criteria

• On-load tap changer: mechanical cycling, contact sequence, transition oscillography

• Insulating oil: breakdown voltage, Karl Fischer water content, physico-chemical characterization

• Bushings: individual C1 and C2 baseline through the test tap

• Temperature rise by total-loss short circuit; the 60, 65 and 78 K limits

• Lightning impulse: the sequence, the comparison method, and reading the neutral current

• Switching impulse on a saturable core and the near-uniform stress distribution

• Short-circuit withstand judged by reactance change; sound level; signature frequency response; dissolved gas analysis

Laboratory work

A directed audit. You receive a complete anonymized routine test report for a 25 MVA, 138/13.8 kV Dyn1 transformer with a 17-position on-load tap changer, containing two planted non-conformities. The routine: (1) reconstruct the timeline of the tests from the record timestamps and check the order against the evidence-preservation algorithm; (2) check the turns ratios of all 17 positions against the nominal ratio and the tap step, applying the ±0.5% tolerance; (3) recompute the temperature correction of the resistances with k = 235 and the I²R versus stray separation of the load losses; (4) verify that declared against measured losses respect the +15% component and +10% total limits and that the impedance respects ±7.5%; (5) check the IVPD profile — levels, duration by the 120 × fn/f formula, pC on the 1.58·Ur/√3 plateau and in the 1.2·Ur/√3 windows — against IEC 60076-3; (6) examine the oil report and the bushing C1 and C2 values against the routine limits, remembering that a condenser bushing baseline is typically a few hundred pF for C1 with tan delta at or below 0.4% for oil-impregnated paper, and that a 3 to 5% rise in C1 in service means a shorted condenser layer. One planted non-conformity is a sequence error, the other is numerical. Whoever finds both is ready to witness a factory acceptance test as the purchaser's technical representative.

Key takeaway

A 30 MVA, 138 kV unit passed every routine test and failed 40 seconds after first energization: differential operation, gas relay, high acetylene in the oil. Reopening the factory report, the sequence was correct, the losses were inside the guarantees, IVPD sat at a stable 80 pC and the impulses were clean. One number was out of line: excitation current measured after the dielectric tests, in the reference repetition, came back 18% higher than the initial reading. It was still inside the contractual tolerance against the declared value, so nobody stopped. But excitation current that grows between before and after is not tolerance, it is an incipient shorted turn — an inter-turn failure started by the impulse, with a contact resistance still too high to disturb the oscillogram, invisible to a ratio measurement because it is a fraction of a turn out of thousands, and silent under reduced-voltage PD. At full voltage on energization, the short welded itself closed. The lesson: tolerances apply against the declared value; before-and-after comparisons apply against the object itself, and they have no tolerance, only trend. A report that treats the reference repetition as paperwork throws away the decisive witness.

Module 12 of the Atlas Energy Academy · Equipment Families · about 8 hours · Video series (12 × 12–18 min) plus factory and laboratory practice · Prerequisites: Module 11 — The Grammar of Testing: Taxonomy and Cross-Cutting Tests