A transformer runs hot for five reasons: the cooling stopped, the load passed the plate, the room went above the design ambient, the load is full of harmonics, or something inside is already failing. In that order of likelihood. The temperature tells you which one before you pick up a wrench, and the plate tells you which of them you are allowed to blame.
Most overheating pages give you a list of causes and stop. That list is not wrong. It is just unusable at 2 a.m. when the winding temperature indicator is climbing and someone has to decide whether to shed load, call an electrician, or do nothing until Monday.
This page is built around that decision. Every cause gets a number that separates it from the other four, a measurement you can take this week, and a fix that does not require replacing the transformer.
One thing before the list. There is already a page on this site that decodes the cooling class letters and explains how a dual rating like 1500/2000 kVA is bought and tested: transformer cooling classes explained: ONAN, ONAF, AN and AF. That page answers what the four letters mean. This one starts later, when the letters are already on the plate and the unit is running hot anyway.
The five causes, and the signature each one leaves
Here is the whole thing on one screen. The middle column is what makes this useful: each cause moves the temperature in a different pattern, and the pattern is more diagnostic than the absolute number.
| Cause | Signature in the readings | First thing to check | Fix |
|---|---|---|---|
| Cooling lost: fan, pump, closed valve, low oil | Hot at any load, and hot out of proportion at light load | Fan rotation, radiator valve position, oil gauge against the temperature chart | Restore the cooling path |
| Load above the plate | Rises with demand, tracks the production shift or the afternoon | Clamp meter reading against nameplate current | Shed, rebalance, or resize |
| Room or ambient above the design basis | Climbs through the afternoon, falls back overnight | Air temperature at the transformer intake, not at the wall | Airflow first, then load limit |
| Harmonics | Hot with phase amps that look perfectly fine | Neutral current and current THD | K-rated unit, filter, or derate |
| Internal fault | Hot and climbing with a flat load, no external reason | Dissolved gas analysis and winding resistance balance | Test, then plan replacement |
The order matters. In our own quote requests, lost cooling and room airflow together account for more overheated units than genuine overload does, and both are cheaper to fix than a new transformer. Overload is the cause buyers assume first and confirm last.
The last row is the one to take seriously. Four of these five causes are reversible. A failing winding is not, and it looks exactly like the other four until you test.
Read the temperature before you read the cause
Three different temperatures get called “the temperature,” and they are not interchangeable. Mixing them up is how a healthy unit gets condemned and a dying one gets ignored.
Top oil, average winding, and hot spot
Top oil is what the gauge at the tank top reads. It is the hottest oil in the unit, and it lags the load by twenty to forty minutes because oil has thermal mass.
Average winding temperature is the copper, measured by resistance. It runs above top oil by the oil-to-copper gradient, typically 15 to 25 K at full load on a distribution unit.
Hot spot is the hottest square inch of insulation anywhere in the winding, usually near the top of the inner winding where the oil is slowest. You cannot measure it directly without fiber optics. It is calculated from top oil, load current, and the manufacturer’s thermal model, and it is the only one of the three that predicts insulation life.
The relationship on a standard 65 °C rise unit is simple arithmetic. ANSI/IEEE rates that unit on a 30 °C average ambient, a 65 K average winding rise, and an 80 K hot spot rise. So the hot spot at full load in a 30 °C room is 110 °C, and that is the number everything else in this article is measured against.
What a 10 K step actually costs
Thermal aging follows the Arrhenius relation, and IEEE C57.91 turns it into an aging acceleration factor. Set 110 °C to 1.00 and the rest falls out:
| Hot spot | Aging factor | What it means in service |
|---|---|---|
| 95 °C | 0.20 | About a fifth of the design aging rate |
| 110 °C | 1.00 | The reference. Design life at rated load |
| 116 °C | 1.83 | Nearly double |
| 120 °C | 2.71 | Life consumed almost three times as fast |
| 130 °C | 6.98 | Seven years of life per year of running |
| 150 °C | 40.6 | A week at this temperature burns about nine months |
The curve is not linear and that is the point. Ten degrees above the reference is not a ten percent penalty. It is a 171 percent penalty.
One warning on the reference point itself. Everything above uses the ANSI/IEEE basis, where the reference hot spot is 110 °C. IEC 60076-7 works from 98 °C and a 6 K doubling interval instead. A unit built and tested to IEC, read against the IEEE curve, looks like a crisis while it is running exactly as designed. Check which standard the plate cites before you convert a reading into a verdict.
Why a 55 °C rise unit is not a 65 °C rise unit with spare margin
Two 1000 kVA transformers, same core, same copper weight on paper. One is rated 65 °C rise, the other 55 °C rise. At full load in the same 30 °C room, the first sits at a 110 °C hot spot and the second at 95 °C.
Using the table above, that is an aging factor of 1.00 against 0.20. The 55 °C unit consumes life about five times slower at the same load. It costs more because it has more copper and more radiator, and the price gap is usually smaller than the life gap.
This is the single most under-specified line in a transformer RFQ, and it is the reason two quotes for the same kVA can differ by thirty percent and both be honest. Check the temperature rise field on the plate before you compare anything else, which is one of the checks in how to read a transformer nameplate and technical parameters.
The 40 °C room that is already too hot
Here is the part that catches people in July. The 110 °C reference assumes a 30 °C ambient. Many North American units are also rated for a 40 °C maximum ambient, and those are not the same statement.
At 40 °C ambient, an unchanged 80 K hot spot rise puts the winding at 120 °C. Nothing has failed. The load is at nameplate. The cooling system is working. The unit is simply aging at 2.71 times the design rate, because the room is ten degrees warmer than the reference.
That is why a transformer can be “running hot” and perfectly healthy at the same time, and why the first question in any overheating investigation is what the ambient was, not what the gauge said.
The 60-second triage: put temperature against load
Before opening a panel, plot two numbers you probably already have: top oil temperature and load as a percentage of nameplate. Where the point lands tells you which section to read.
| Reading | Most likely cause | Why |
|---|---|---|
| Hot at full load, normal at 50 percent | Load, ambient, or a design that was always marginal | Heat tracks load squared, so this is the expected shape |
| Hot at 50 percent load or below | Cooling has failed, or the reading is wrong | At half load the unit rejects about 64 percent less heat than at full load |
| Normal all winter, hot all summer | Room airflow or ambient basis | Follows weather, not production |
| Hot on one phase only | Unbalanced load, or a bad connection | Two healthy phases rule out the cooling system |
| Hot with all three phase amps equal and low | Harmonics, or an internal fault | Current is not telling the story, so something else is |
The second row is the one people get wrong. A 1000 kVA unit with 1,500 W of no-load loss and 9,000 W of load loss rejects 10.5 kW at full load. At half load the copper loss falls to a quarter, so total heat is 1,500 plus 2,250, or 3.75 kW. That is 64 percent less heat than at full load.
If a transformer with that loss profile is hot at half load, overload is not your problem and no amount of load shedding will fix it. Go to the cooling system.
Cause 1: the cooling system quit and nobody noticed
This is the most common real cause and the easiest to confirm. It also has the largest immediate consequence, because losing cooling does not just make the unit hot. It removes capacity the nameplate says you have.
Fans: the 133 percent that disappears
A dual-rated 1500/2000 kVA unit is 2000 kVA with the fans running and 1500 kVA with them stopped. That is 133 percent, and the step down is instant.
Put 1,800 kVA of load on it, which is 90 percent of the top rating and a comfortable number on a mild day. Lose the fan bank, and the same load is now 120 percent of the 1500 kVA ONAN rating.
The heat does not change when the fans stop, and that is the part people get backwards. With 2,200 W of core loss and 17,500 W of load loss at 2000 kVA, the unit at 1,800 kVA generates 2,200 W plus 14,175 W, or 16.4 kW, whether the fans are turning or not. What changes is the capacity to reject it. In ONAN mode the same radiators can only carry about 12.0 kW while staying inside the rated temperature rise, because that is what the 1500 kVA rating means.
So the tank is being asked to shed 36 percent more heat than the ONAN cooling arrangement can handle. That is a unit that goes from fine to alarming in an afternoon, and the only thing that changed is a contactor.
Practical check: walk the fans while the unit is loaded. Do not trust the controller. A fan with a burned motor and a free-spinning blade sounds like a fan. Look for the ones that are not turning, and confirm the blades are clean enough to move air rather than just spin.
Pumps, and the OFAF units
Above roughly 20 MVA, and on many units from 5 MVA up in industrial service, oil pumps add a forced circulation stage. A pump failure is quieter than a fan failure because there is nothing to see. Oil simply stops moving through the radiators, and the oil-to-air gradient climbs while the top oil reading stays deceptively normal for a while, since the gauge is sitting in a stagnant pocket at the top of the tank.
On forced-oil units, check differential pressure across the pump and the flow indicator, not just the run status. A pump can be turning and moving nothing.
Radiator valves, fins, and the 11 K valve
Radiator banks are isolated by valves so a leaking radiator can be removed without draining the tank. Those valves get closed for maintenance and stay closed, because nothing in the building couples “valve closed” to “capacity lost.”
The arithmetic is unforgiving. Take a six-bank radiator set designed for a 55 K oil-to-air gradient at full load. Close one bank and the remaining surface has to reject the same heat, so the gradient rises in proportion to the area lost, from 55 to 66 K. That is 11 K added to the top oil, and 11 K added to the hot spot.
At 110 °C, an 11 K rise moves the aging factor from 1.00 to about 2.99. One valve, closed by hand, and the unit burns three years of insulation life per year of operation. This is the cheapest fault in this article to fix and among the most expensive to leave.
Fins matter for the same reason. A radiator clogged with cottonwood, dust, or a bird nest behaves like a closed valve. On outdoor units in agricultural or construction dust, annual washing is not cosmetic maintenance, it is capacity maintenance.
Low oil level
Oil is the heat path from copper to tank wall. Drop the level below the top radiator inlet and the top of every radiator stops working, which is the same failure as closing a valve except you cannot see it and the leak is still running.
Read the magnetic oil gauge against the temperature chart on the tank, not against a mark someone painted. Oil volume changes roughly 0.07 percent per degree Celsius, so a unit that reads low at 10 °C may read normal at 60 °C. A gauge that is low when cold and normal when hot is not a gauge that is fine. It is a gauge that is low.
Cause 2: load above the rating
Load loss rises with the square of current, which is the only equation you need for the rest of this section.
At 120 percent load, copper loss is 1.44 times the full-load value. At 150 percent it is 2.25 times. At 200 percent it is four times. The heat curve has no soft shoulder.
For a 1000 kVA unit with 9,000 W of load loss, 120 percent load means 12,960 W of copper loss on top of 1,500 W of core loss. Total heat goes from 10.5 kW to 14.5 kW, a 38 percent increase in what the cooling system has to reject, while the cooling system itself is sized for the 10.5.
What the load actually is
Nameplate current for 1000 kVA at 480 V is 1,203 A. That is the number to compare against a clamp reading. It is also the number most often compared against the wrong measurement: the sum of breaker ratings on the panel schedule, which is a design maximum and always larger than the real load.
Three things inflate real load beyond what the schedule says: motor starting inrush (short, thermal mass absorbs it), EV charging and data loads with poor power factor (current stays high while kW looks modest), and single-phase loading on a three-phase unit, which heats one winding more than the average suggests.
Overload that is legal and still costs you
The loading guides permit planned overload. IEEE C57.91 and IEC 60076-7 both let you run above nameplate for a defined period at a defined ambient, in exchange for a defined amount of insulation life.
That is a real engineering option during a heat wave or a temporary outage, and it is booked as a cost, not free capacity. The trade is explicit: an extra 10 K of hot spot for a week costs roughly the equivalent of a couple of months of normal aging, and the guides expect you to record it.
What the guides do not permit is running at 120 percent for a year and calling it a design margin. By the table above, that is roughly 2.7 years of life per calendar year, so a 30-year unit becomes an 11-year unit without a single alarm.
If your load has crept up 20 percent over five years through small additions, you are in this section whether or not anyone decided to be there.
Cause 3: the room, the ambient, and the altitude
A transformer rejects heat to air it does not own. If the room cannot carry that heat away, the unit overheats no matter how healthy it is, and no electrical fix applies.
The airflow number nobody checks
Room ventilation is sized from losses, not from kVA. The relation is a standard sensible heat balance:
CFM = 3.16 x watts of loss / allowable room rise in deg F
m3/s = watts of loss / (1200 x allowable room rise in deg C)The constant 3.16 is just air density times specific heat, folded into imperial units. It assumes sea level and standard air. It does not care what the equipment is.
Work it on three common sizes, using 15 °F (about 8.3 °C) of allowable room rise, which is a typical design target:
| Rating | No-load loss | Load loss | Total heat at full load | Airflow needed at 15 °F rise |
|---|---|---|---|---|
| 500 kVA | 900 W | 5,200 W | 6.1 kW | About 1,290 CFM |
| 1000 kVA | 1,500 W | 9,000 W | 10.5 kW | About 2,210 CFM |
| 2000 kVA | 2,200 W | 17,500 W | 19.7 kW | About 4,150 CFM |
Two thousand CFM is not a vent. It is a mechanical exhaust system with a fan, a controller, and a maintenance schedule, and it is what a 1000 kVA dry-type or oil unit needs if it sits in a sealed room.
Louver area follows from airflow. A working rule is one square foot of free area per 300 CFM. For 2,210 CFM that is about 7.4 square feet of free area, split between a low intake and a high exhaust. A weather louver passes roughly half its face area as free area, so the opening you cut in the wall is closer to 15 square feet per side.
The exhaust goes high, because that is where the heat is. The intake goes low, because cool air enters at floor level and does the work on the way up. A room with both openings at the same height, or with the intake above the transformer, is a room that will run hot forever.
40 °C ambient is the ceiling, not the average
Two different ambient numbers appear on nameplates and they get used interchangeably. A 30 °C average ambient is the basis for the temperature rise rating. A 40 °C maximum ambient is the limit at which the rating still holds.
A rooftop in Phoenix, a boiler room, a mezzanine above a process line, and a shipping container substation all spend summer above 40 °C at the intake. When that happens the unit is outside its rating basis and the options are three: add airflow to bring the intake down, shed load to bring the hot spot back to 110 °C, or accept accelerated aging with a date on it.
Adding airflow is usually the cheapest of the three and the one nobody quotes. Measure the intake temperature at the transformer, at 2 p.m., in August, before you conclude anything. A room reading of 38 °C at the wall and 47 °C at the intake is a room with an airflow problem, not a temperature problem.
Altitude
Air at 2,000 m carries roughly 80 percent of the mass of air at sea level in the same volume. Fans move the same volume and therefore less mass, and natural convection gets weaker in the same proportion. Both the 1,000 m reference in IEC 60076-11 and the manufacturer’s altitude correction exist because of this.
If a unit is specified at sea level and installed at 2,000 m, its effective cooling is reduced and its rating is not the nameplate value. This shows up as a unit that is “undersized” in a mountain or high-plateau installation where the math said it was fine. Ask for the correction table with the quote, not after commissioning.
The enclosure trap
An outdoor dry-type in a NEMA 3R enclosure is sold as weather-protected and it is. What it is not is ventilated the way an open indoor unit is. If the enclosure has filters, the filters clog. If it has louvers, the louvers face the sun on one side.
Enclosed outdoor units need their airflow path inspected on a schedule, which is why outdoor dry-type transformer options for exposed sites are usually specified with a larger thermal margin than the indoor equivalent. The margin is buying you the privilege of a dirty filter.
Cause 4: harmonics heat it without moving the amps
A transformer serving variable frequency drives, rectifiers, LED drivers, servers, or anything with a switch-mode supply carries current that is not a sine wave. The fundamental component may be well inside the rating. The harmonic components are not free.
Harmonic currents increase three loss mechanisms at once. They raise the RMS current above the fundamental, so I squared R goes up. They raise eddy current loss in the windings, and that loss rises roughly with the square of frequency, so the fifth harmonic contributes twenty-five times the eddy loss per amp of the fundamental. They raise stray and structural loss in the tank, frames, and core clamps, which is where local hot spots form that the top oil gauge never sees.
This is why a unit can be at 70 percent of nameplate current, cool oil, and still cooking a winding.
Triplen harmonics and the neutral
Third, ninth, and fifteenth harmonics add in phase in a four-wire system instead of cancelling. On a 480Y/277 V unit serving office or data loads, the neutral can carry more current than any phase, and it is often not counted in the load calculation at all because the schedule shows three phase amps and a neutral.
If the neutral is hot enough to be uncomfortable to touch, the transformer is being loaded in a way the kVA number does not describe, and the winding is hotter than the current suggests.
K-factor, and what it actually buys
A K-rated transformer is built with a larger neutral, transposed winding conductors to control eddy loss, and a core and clamp design that tolerates stray flux. The K number says how much harmonic content it can carry at full nameplate kVA without exceeding its temperature rise.
If your load is more than about 15 percent nonlinear by kVA and you are buying a standard unit, you have two honest options: buy a K-rated unit, or derate a standard one and accept that the derating factor is an estimate. The usual practical shortcut of “add 25 percent kVA” is a guess that happens to be right for moderate drive loads and wrong for dense rectifier loads.
Data centre and drive-heavy industrial loads live in this section permanently, which is why data centre transformer solutions are usually specified around the harmonic spectrum rather than the kVA alone.
Cause 5: internal faults wearing a thermal disguise
Everything above is external. This section is not, and the difference matters because the fix is a purchase order rather than a service call.
Circulating currents and shorted laminations
A few shorted laminations in the core form a closed loop that carries circulating current. The loss is small in absolute terms and extremely local, so it produces a hot spot in the core that heats the oil around it without raising the average temperature much.
The signature is a unit that runs a few degrees hotter than its twin on the same bus, with normal current, normal power factor, and no external explanation. Winding resistance balance and a core-to-ground insulation test will separate this from a cooling problem.
Sludge, blocked ducts, and oil that stopped being oil
Oil oxidizes. Oxidation products polymerize into sludge, and sludge collects where the oil moves slowest, which is inside the winding ducts and at the bottom of the tank. Blocked ducts do not reduce the total oil volume, so the gauge reads fine while the oil that should be cooling the inner winding cannot get there.
The signature is a rising top oil temperature at constant load over months, plus DGA showing carbon oxides climbing faster than hydrocarbon gases. That is cellulose being cooked, not oil being decomposed, and it is the pattern that says the insulation is being consumed from the inside.
Once sludge has formed, oil processing removes dissolved products but does not reopen a blocked duct. This is where the decision turns into replace versus rewind, and where a routine oil test program pays for itself. The test set that catches it early is the same one in transformer routine testing: what every unit should pass.
Cooling methods, and what each one actually buys
Cooling classes are a ladder, and each rung buys capacity with hardware that can fail. That trade is the whole design decision.
The four-letter ladder
The code on the plate reads in four positions: internal coolant, internal circulation, external medium, external circulation. ONAN is oil natural, air natural. ONAF adds fans. OFAF adds pumps. ODAF directs pumped oil through the winding ducts to attack the hot spot directly.
| Designation | What it adds | Where you meet it | What takes it away |
|---|---|---|---|
| ONAN | Nothing. Buoyancy and radiation only | Distribution, up to roughly 10 MVA | A shut valve, a fouled bank |
| ONAF | Fans blowing across the radiators | 5 to 60 MVA, most industrial units | One fan bank, clogged fins |
| OFAF | Oil pumps plus fans | 20 MVA and up, and 5 MVA and up in heavy industry | Pump loss, low flow, blocked strainer |
| ODAF | Pumped oil directed through winding ducts | The largest power transformers | Blocked directed-flow paths |
| AN (dry-type) | Room air across the winding surface | Indoor dry-type, commercial | Room airflow, enclosure filters |
| AF (dry-type) | Fans across the winding | Dual-rated indoor dry-type | Fan failure, dust on the windings |
The first step up buys about a quarter to a third more capacity from the same core and copper. Later steps buy less, because oil flow stops being the bottleneck and the winding gradient starts to be. Every step costs auxiliaries that need power, maintenance, and a failure plan. The full decode, including the K and A prefixes and the dry-type equivalents, is on the cooling class page linked at the top of this article.
The engineering point that matters here: the top rating on a dual-rated unit is not a bigger transformer. It is the same transformer with working fans. That is why a fan outage is a capacity event and not a comfort issue.
Dry-type: AN, AF, and the 150 K rise
Dry-type units run hotter by design and it is not a fault. The common North American specification is a 220 °C insulation system at a 150 °C rise, which at a 40 °C ambient puts the winding near 190 °C at full load.
An oil unit next to it is at 110 °C. Both are normal, and comparing their readings is how people talk themselves into a false alarm or, worse, into ignoring a real one on the oil unit because “the dry-type always reads higher.”
Dry-type ratings also carry an insulation class letter. IEC 60076-11 uses class F at 155 °C and class H at 180 °C, with corresponding rise limits at a 40 °C maximum ambient. The margin between the class temperature and the operating temperature is what absorbs overload, and it is why a Class H unit costs more for the same kVA.
Forced air (AF) on a dry-type does what ONAF does on an oil unit: it buys a second rating. Same caveat applies, and it is louder, because there is no oil to damp the fan noise.
Water cooling, and the failure nobody plans for
OFAF and OFWF units reject heat to water through a heat exchanger. The thermal performance is excellent and the failure mode is different from anything else in this article: a water-side blockage or a lost cooling water supply removes most of the capacity in minutes, not hours, because the oil has nowhere to dump heat.
If a water-cooled unit is part of your plant, the cooling water supply belongs on the same critical-load list as the transformer itself. Including the strainer cleaning.
The retrofit trap: fans do not change the nameplate
Adding fans to an ONAN transformer is a real practice and it does something useful. It lowers the hot spot at the same load, which extends life, and it gives you thermal headroom for occasional peaks.
What it does not do is change the rating. The higher kVA figure has to come from a type-tested cooling arrangement, because the manufacturer’s thermal guarantee is written against a tested configuration and nobody can measure the winding hot spot in the field to prove the uprate worked.
So the honest version is: field fans buy margin and life, not kVA. Buying them as a capacity increase is a decision that works until the first insurer or utility review asks for the test report.
Measuring it honestly
You cannot manage a temperature you are measuring wrong, and most overheating arguments are measurement arguments in disguise.
OTI and WTI, and the resistor nobody set
The oil temperature indicator is a bulb in a well at the tank top. It reads oil, it lags load by twenty to forty minutes, and it is reliable.
The winding temperature indicator is an OTI plus a small current-driven heater that simulates the winding gradient on top of the oil temperature. The heater current comes through a gradient resistor, and that resistor is set at commissioning to match the measured top-oil-to-hot-spot gradient for that specific unit.
If the unit was never heat-run, or the resistor was left at a default, the WTI reads a plausible number that has no relationship to the actual hot spot. It can be 10 to 15 K off in either direction, which is the difference between “normal” and “shut it down.” Ask for the commissioning record. If there is not one, treat the WTI as an indicator and not as an instrument.
Why infrared cannot see the hot spot
Infrared surveys are excellent and they are not a winding temperature measurement. On a radiator-cooled unit the tank surface sits somewhere around 45 to 60 °C while the winding hot spot is at 110 °C, because the radiators are doing the work and the tank wall is not where the heat is.
What infrared finds is what happens outside the tank: loose bus connections, failing bushings, cable terminations, control wiring, and cooling fan motors. Those are worth finding. They are also not the winding.
The unit that has a genuine internal hot spot and a normal tank surface is exactly the unit that gets signed off by an infrared survey. This is the most common false negative in transformer condition assessment.
The delta-T ladder, and the standard that made it mandatory
For the connections and bushings infrared does cover, the severity grading comes from NETA Maintenance Testing Specifications Table 100.18. NFPA 70B requires you to measure and document the temperature difference but does not itself set the severity bands, so the ladder in practice is NETA’s:
| Delta-T against a similar component | Delta-T against ambient | Priority | Action |
|---|---|---|---|
| 1 to 3 °C | 1 to 10 °C | 4 | Possible deficiency, recheck next round |
| 4 to 15 °C | 11 to 20 °C | 3 | Probable deficiency, repair as scheduled |
| Not applicable | 21 to 40 °C | 2 | Monitor until corrected |
| Above 15 °C | Above 40 °C | 1 | Major, correct immediately |
Two caveats that decide whether the report means anything. First, the scan needs load: a connection only heats when current flows, and industry practice calls for at least 40 percent of normal load, with peak load better. Second, delta-T scales roughly with current to the power of 1.6 to 2, so a reading taken at half load understates the full-load condition by a factor of about four and must not be compared against the table as if it were a full-load number.
NFPA 70B moved from a recommended practice to a standard in the 2023 edition and now requires documented infrared inspection on a defined interval, annually for equipment in normal condition and more often for equipment already flagged. The 2026 edition is the current one. If your insurer has started asking for survey records, this is why.
Prevention: a calendar with numbers
Prevention here is not a culture program. It is six tasks with intervals and pass criteria, and every one of them is cheaper than the failure it prevents.
| Interval | Task | Pass criterion |
|---|---|---|
| Quarterly | Run-test every fan bank and verify the control stages | Each bank starts on command and on temperature setpoint |
| Quarterly | Read top oil against load and log both | The pair tracks the same curve as last quarter |
| Annually, before summer | Wash radiator fins and check every isolating valve | Valve open, fins clear through the bank |
| Annually | Cross-check OTI and WTI against a contact thermometer | Within 3 K at the tank top |
| Annually | Oil test, including DGA and moisture | No gas trending up at constant load |
| Annually, at 40 percent load or better | Infrared survey of terminations and bushings | No priority 1 or 2 findings outstanding |
| Each August, at peak | Measure intake air temperature at the transformer | Below the ambient basis on the plate |
| Every 3 years | Audit cooling stage setpoints against the thermal model | Setpoints match the current load profile |
The two lines that catch the most real faults are the cheapest. Valve position costs nothing and accounts for the 11 K example above. Intake temperature costs a five dollar thermometer and explains most July complaints.
What to put in the RFQ so this does not happen
Most overheating stories start with a quote that never mentioned temperature rise, cooling stages, or ambient. Here is the block we ask buyers to fill in, and the block we return filled in:
Rated capacity and cooling stages: 1500 kVA ONAN / 2000 kVA ONAF, both type tested
Temperature rise: 65 K average winding, 80 K hot spot, 30 C basis
Site ambient and altitude: 40 C maximum, 1200 m elevation, indoor room
Cooling hardware: 6 fans in 2 banks, N+1 bank, valves lockable open
Monitoring: OTI and WTI, 2 alarm + 1 trip contacts, 4 x RTD
Losses: P0 and Pk guaranteed values, test report supplied
Harmonic duty: K-factor 13, neutral sized for 200% triplen
Efficiency basis: DOE 10 CFR 431 subpart K, 2029 tableTwo lines do most of the work. The first forces the vendor to state whether the top rating is tested or advertised. The seventh decides whether the unit will still be cool in year ten on a drive-heavy bus.
The efficiency line is not decoration either. The 2029 DOE loss tables under 10 CFR 431 subpart K cap losses, and capped losses are capped heat, which is why a compliant unit runs cooler than the 2016-vintage one it replaces at the same load. If you are replacing a hot unit, checking the loss guarantee is the only way to know whether the replacement will actually run cooler or just be newer. We size and quote those units against the 2029 table by default, and the spec sheet for S20 oil-immersed industrial transformers lists the loss values to compare against.
Send the plate photo and the room dimensions and we will come back with the airflow number, the cooling stages that fit, and a loss guarantee you can put next to the quote you already have. The contact form takes all of it in one upload.
FAQ
What is a normal operating temperature for a transformer?
For an oil-immersed unit rated 65 °C rise, normal is a top oil around 85 to 95 °C at full load in a 30 °C room, with a winding hot spot near 110 °C. Below 100 °C top oil you are inside the design basis. Between 100 and 105 °C you are using life faster than design. Above 105 °C at rated load, something in the cooling path has failed.
At what temperature should a transformer be shut down?
Common settings are a top oil alarm near 95 °C and a trip near 105 °C, with a winding hot spot alarm at 110 °C and a trip between 130 and 140 °C. These are starting points, not universal values. The correct setpoints come from the manufacturer’s thermal model for that specific unit, and they should be verified at commissioning rather than assumed.
Can a transformer overheat at half load?
Yes, and it is the clearest sign that the problem is not overload. At half load the copper loss falls to a quarter of its full-load value, so a 1000 kVA unit with 10.5 kW of total loss at full load rejects only about 3.75 kW. If it runs hot there, cooling has failed, a radiator valve is shut, the oil is low, or the reading is wrong.
How much airflow does a transformer room need?
Size it from losses, not kVA. The relation is CFM equals 3.16 times watts of loss divided by the allowable room rise in Fahrenheit. A 1000 kVA unit with 10.5 kW of loss needs about 2,210 CFM for a 15 °F room rise, and a 2000 kVA unit with 19.7 kW needs about 4,150 CFM. Louver free area runs about one square foot per 300 CFM.
Does adding fans increase a transformer’s kVA rating?
Not legally. Fans lower the hot spot at the same load and buy thermal margin, but a higher kVA rating has to come from a type-tested cooling arrangement with the manufacturer’s guarantee behind it. Field-installed fans void that guarantee because the winding hot spot cannot be measured in service to prove the uprate. Buy the rating, then buy the fans.
Why does my transformer run hotter in summer than in winter?
Two effects stack. Ambient air is warmer, so the same temperature rise lands on a higher absolute hot spot, and air is less dense when hot, so a given fan moves less mass and the radiators reject less heat. At 40 °C ambient a 65 °C rise unit sits at a 120 °C hot spot at full load, which is about 2.7 times the design aging rate with nothing actually broken.
How much life does running 10 °C hot cost?
About 170 percent more aging, not 10 percent. On the IEEE C57.91 aging curve, 110 °C is the 1.00 reference and 120 °C is 2.71, so a year at the higher temperature consumes roughly 2.7 years of insulation life. At 130 °C the factor is 6.98, and at 150 °C it is 40.6, where a week costs about nine months.
Can a thermal camera find an overheating winding?
No. On a radiator-cooled unit the tank surface runs around 45 to 60 °C while the winding hot spot is near 110 °C, because the radiators reject the heat and the tank wall is not the hot path. Infrared finds loose connections, failing bushings, and fan motor problems. Internal winding heat is only visible through the oil and winding temperature indicators and through dissolved gas analysis.
What can cause overheating with no overload at all?
Four things that have nothing to do with load: a closed radiator isolating valve, a failed fan bank on a dual-rated unit, a fouled radiator or blocked room louver, and harmonics from nonlinear loads. A fifth is internal, either shorted core laminations or sludge blocking winding ducts. The first four are visible on a walk-around in under an hour.
Is harmonic distortion really a heating problem?
It is one of the most underestimated ones. Harmonic currents raise RMS current, raise eddy loss roughly with the square of frequency, and raise stray loss in the tank and clamps where local hot spots form that the top oil gauge never sees. A unit can sit at 70 percent of nameplate current, show normal oil temperature, and still be cooking a winding. Triplen harmonics also load the neutral, which most kVA calculations ignore.
