Explainer
Evaluating dry-type transformer hot spot temperature rise
Short answer
A 128.2°C average winding rise proves only the average, not the hot spot: the resistance method and a point RTD measure different quantities, so the 219°C combination is invalid. The 202.6°C full-load reading is a 180.8°C local rise, which projects to 220.8°C at the 40°C rating ambient and exceeds the 180°C hot-spot rise a class 220 system allows. Decision: run the simultaneous full-voltage, full-load heat run.
On this page
The number that settles this case is 220.8 °C. The full-load RTD reading of 202.6 °C, taken at a 21.8 °C test ambient, is a local rise of 180.8 °C; projected to the 40 °C ambient a class 220 rating is built on, it reaches 220.8 °C and crosses the 220 °C insulation limit. The manufacturer's 128.2 °C average winding rise is a valid result against the 150 °C design, but the resistance method measures a winding average and cannot bound a local hot spot, and the 219 °C figure produced by feeding RTD values into the average-rise formula is not evidence of anything. Decision: do not accept the transformer on these numbers, and run the simultaneous full-voltage, full-load heat run — the only test mode that puts core loss and load loss in the winding at the same time.
What class 220 insulation actually allows #
Class 220 is a hot-spot limit for the insulation system, not an average-winding limit. IEEE C57.12.01, the IEEE Standard for General Requirements for Dry-Type Distribution and Power Transformers (2020 edition), covers the electrical, mechanical and safety requirements for ventilated dry-type units together with the temperature-rise test methods used to prove them; IEC 60076-11:2018 defines the equivalent temperature classes in the IEC framework. The rating arithmetic behind a class 220 unit with the usual 150 °C rise design is:
220 °C hot-spot limit = 40 °C maximum ambient + 150 °C average winding rise + 30 °C hot-spot allowance
Read in reverse, the allowable hot-spot rise is 180 °C at a 40 °C ambient, and it rises 10 °C for every 10 °C the ambient sits below that, giving 190 °C at the plant's maximum room temperature. The usual service condition behind the rating is an ambient that never exceeds 40 °C, so 40 °C is the ambient an acceptance screen must use unless the nameplate and purchase specification say otherwise. The units in this case were built in 1996 against the 1989 edition of that standard, so confirm the governing edition, limits and correction methods with the test laboratory, the transformer documentation and the authority having jurisdiction, in writing, before any witnessed run.
The evidence from the failed line-up #
The failed unit was a 12 MVA, 34.5 kV to 4.16 kV ventilated dry-type transformer, 150 °C rise, class 220, built in 1996, feeding a 4.16 kV pump-motor lineup through a fused primary switch and a secondary main breaker. It failed as an interturn secondary-winding fault, with no ground on either winding afterward; teardown found the low-voltage turns nearest the core baked and flaking, worst at the top of the center winding.
Nothing in the operating record points to overload: 7.7 to 10.8 MVA sustained, a maximum 15-minute average just under 9 MVA (75% of rating), primary voltage at or below 103%, voltage THD 1.3%, current THD under 1%, room ambient at or below 30 °C. No heat run had been performed at manufacture. Infrared thermography on the two siblings from the same purchase read the core above the center winding at 220 °C, up to 239 °C. The 10 MVA sibling then went to the manufacturer for a witnessed heat run:
| Test condition | Ambient | Core center leg, 1 in. from winding top | LV RTD, first wrap, 15 in. from coil top | LV average rise by resistance |
|---|---|---|---|---|
| 111% excitation, no load | 24 °C | 186 °C steady state | 98 °C | 37.9 °C |
| Secondary shorted, full-load current | 21.8 °C | 152 °C steady state | 202.6 °C steady, 209 °C peak | 105.2 °C |
The manufacturer combined the two resistance-derived average rises into 128.2 °C under the standard's rule for merging separate-loss test results, which passes the 150 °C average criterion. Every other number in that table is a local temperature.
Average rise and hot spot are different quantities #
The resistance method converts the change in winding resistance, measured on the cooling curve immediately after shutdown, into one temperature for the whole winding: an average. An RTD reports one point. The hot-spot allowance in the rating arithmetic is a design estimate of how far the hottest point sits above that average, and it holds only where the loss distribution and the airflow match the design.
Winding ends carry a radial flux component that drives eddy losses in the conductor; tight core-to-coil clearances and blocked ventilation concentrate heating locally; and in air-cooled dry-type windings the hottest spot sits further above the average than in liquid-immersed units — measurements on naturally ventilated dry-type coils go back to 1944 and have always shown the larger spread. A passing average with cooked insulation at the top of one winding is therefore not a contradiction; it is the signature of a local problem the average cannot see.
Why the 219 °C calculation proves nothing #
Three independent reasons, any one of which is fatal:
- The formula merges resistance-derived average rises. Substituting point readings — 98 °C and 202.6 °C — changes the quantity being calculated; a location temperature is not a winding average, and nothing in the method establishes that the formula applies to local values.
- The two tests drive different loss sets. The 111% excitation test drives core loss with no load current; the short-circuit test drives full-load current through the windings with almost no excitation. Each test alone understates the simultaneous condition, and local temperatures from the two modes cannot be added into it.
- The sequence contaminates the peak. Excitation ran first and load second, so the 209 °C peak rode on residual core heat while the winding was still climbing — a transient overlap, not a steady hot spot.
Discard the 219 °C figure from every acceptance argument, on either side of the table.
The ambient-corrected screen that does work #
A defensible screen keeps each measured local rise intact and adds the rating ambient:
projected temperature = measured reading − test ambient + rating ambient
The assumption — that the local rise is unchanged when ambient changes — is stated here, so each result below is a screening projection to be settled by the acceptance test, checked against the governing standard and the transformer documentation, not a substitute for it.
| Reading | Local rise | Projected at 40 °C rating ambient | Projected at the site maximum ambient |
|---|---|---|---|
| 98 °C no-load RTD | 98 − 24 = 74 °C | 114 °C | 104 °C |
| 202.6 °C steady load RTD | 202.6 − 21.8 = 180.8 °C | 220.8 °C | 210.8 °C |
| 209 °C peak load RTD | 209 − 21.8 = 187.2 °C | 227.2 °C | 217.2 °C |
Against the 180 °C allowable hot-spot rise at 40 °C, the steady reading is 0.8 °C over and the peak 7.2 °C over. At the plant's maximum room temperature the margins are 9.2 °C and 2.8 °C under the 190 °C allowance at that ambient — thin, for a single RTD on the first, core-side wrap 15 in. from the coil top, which is an assumed hot-spot location and not guaranteed to be the hottest point. The in-service thermography is the tiebreaker: 220 to 239 °C surfaces at no more than the recorded room maximum are local rises of at least 190 to 209 °C against the same 190 °C allowance. Published aging estimates sharpen the stake — one transformer example projects a 20-year life at an 80 °C hot spot collapsing to 10 years at 88 °C — so decades of operation in the 210 to 220 °C range fully explain a 1996 winding failing at 75% load.
Between acceptance tests, routine ambient and terminal-heat checks on dry-type transformers catch the blocked airflow and connection heating that push a marginal unit over its limit.
The test that settles it #
Any sensor move, resistance connection or enclosure opening on these units happens only after de-energizing, locking out, testing for absence of voltage and grounding the primary — NFPA 70E (2024 edition) sets the minimum electrical safe work practices — and the energized heat run is witnessed from outside the test barriers.
The manufacturer's request for a third-party full-voltage, full-load test is technically the correct request, and the cost is justified: it is the only mode that puts excitation loss, I²R loss and stray losses into the winding simultaneously, which is the condition the failed unit lived with. Agree in writing before the run:
- The governing edition: IEEE C57.12.01-2020 methods, and how the 1989-edition criteria this 1996 unit was built under are treated.
- A sensor placement map that includes the previously instrumented locations and enough coverage to search for a hotter point.
- The steady-state criterion and minimum dwell, with continuous time histories retained for every RTD and ambient sensor — a single peak cannot distinguish equilibrium from retained heat.
- The ambient correction and the resistance-rise calculation, fixed before testing rather than after.
- Stated measurement uncertainty for the RTD chain.
- Acceptance limits set in advance: average winding rise against 150 °C, corrected hot-spot temperature against 220 °C.
The decision for each unit #
- The 219 °C combination: discard. Wrong input quantity, mismatched test modes, sequence-contaminated peak.
- The 128.2 °C average rise: accept for what it is. The average criterion passes; it was never a hot-spot result.
- The 10 MVA unit: run the full-voltage, full-load heat run. On the screen above — 220.8 °C at the rating ambient — expect the hot-spot limit to fail unless the simultaneous condition beats the separate tests, and line up the remediation path now.
- The 7.5 MVA sibling: same test. In the interim, derate and instrument: continuous RTD monitoring with alarms set below the 180 °C rise allowance, and a load cap set by the monitored rise rather than a fixed percentage.
- Protection: add transformer differential (87T) and pull the primary timed overcurrent — which covers six substations — back inside the transformer thermal limit. Protection detects the next interturn fault sooner; it does not repair the thermal design.
- The failed 12 MVA: treat the design, not the operation, as the problem. The record — 75% loading, clean power, cool room, baked insulation at the top of the core-side winding — supports a winding that ran locally above its insulation capability for its service life.
Questions asked
Is the 219°C figure from combining the RTD readings valid evidence?
No. The formula merges resistance-derived average winding rises; an RTD reads a single point, so inserting 98°C and 202.6°C changes the quantity being calculated. The two tests also drive different losses — core loss at 111% excitation, load loss with the secondary shorted — and the excitation test ran first, so residual core heat overlaps the 209°C peak. Use it to justify the proper test, never as an acceptance number.
What does the 202.6°C RTD reading become at a 40°C ambient?
Subtract the measured 21.8°C ambient to get a 180.8°C local rise, then add the 40°C rating ambient: 220.8°C, over the 220°C class limit. At the plant's maximum room temperature the same reading projects to 210.8°C, passing the 190°C allowance at that ambient by 9.2°C, and the 209°C peak passes by only 2.8°C.
Does the 128.2°C average rise clear the transformer?
It clears the average-rise criterion only. Class 220 pairs a 150°C average winding rise with a hot-spot allowance, and an average says nothing about whether that allowance holds at any one location. The failed unit's damage at the top of the center winding, and the 220 to 239°C thermography on its siblings, are exactly the local behavior an average cannot see.
Why did the winding fail at about 75% load?
Long-term local overheating, not overload. The unit ran 7.7 to 10.8 MVA with a 15-minute peak under 9 MVA on a 12 MVA rating, harmonics were negligible, voltage stayed at or below 103%, and the room held the recorded site maximum. Aging is steep near the limit — one published estimate halves a 20-year life between an 80°C and an 88°C hot spot — so decades at a marginal local temperature are enough.
What must the full-voltage, full-load test specify before it starts?
Agree in writing: the governing edition of IEEE C57.12.01, the sensor placement map, the steady-state criterion and dwell, the ambient correction, the resistance-rise calculation, the measurement uncertainty, and the acceptance limits — average rise against 150°C, corrected hot spot against 220°C. Retain continuous time histories for every sensor, and treat any single peak as a transient unless it persists at steady state.