Power Transformer Protection: Complete Guide (87T, REF, 24)

A power transformer is usually the most expensive single asset in a substation and the one with the longest replacement lead time. It is also the one item of plant where the protection engineer has to defend against two entirely different classes of failure at once: fast electrical faults that must be cleared in well under 100 ms, and slow thermal or dielectric degradation that develops over hours or years and produces almost no fault current at all. No single relay function covers both, which is why transformer protection is always a layered scheme rather than one clever algorithm.

This guide is the hub for everything on PowerSynchro about protecting power transformers – what each function actually detects, how the functions overlap, where the classic commissioning traps are, and how the pieces fit into one coordinated tripping philosophy. Each section links to a detailed article on that specific function.

1. Transformer Failure Modes and What Detects Them

Protection schemes make far more sense when you start from the failure, not from the relay. A transformer can fail in at least eight distinguishable ways, and they do not all produce fault current.

Two observations from that table drive everything else on this page. First, the two most common serious faults interturn and near-neutral earth faults are exactly the two the differential relay handles worst. Second, several failure modes produce no differential current whatsoever. That is the entire justification for a layered scheme.

2. How Much Protection? Scheme Selection by Rating

There is no universal rule, and utility standards differ, but the following is a reasonable industry-typical progression. Cost of the scheme is trivial next to the cost of the transformer above about 10 MVA, which is why the step change happens there.

3. The Complete Protection Scheme

4. Differential Protection (87T)

Differential protection is the main protection on any transformer above roughly 10 MVA. It compares current entering the protected zone with current leaving it: under healthy conditions and for any external fault, the vector sum of the compensated winding currents is close to zero; for an internal fault it is not.

What makes a transformer differential harder than a busbar or generator differential is that the “healthy” sum is never exactly zero. Four unavoidable sources of standing differential current have to be accommodated:

• CT ratio mismatch: HV and LV rated currents differ by the turns ratio, and standard CT ratios rarely land exactly on the required value.
• On-load tap changer range: a ±10% OLTC introduces up to about 10% spill current at the tap extremes, and settings must hold for the full range, not just nominal tap.
• Vector group phase shift: a Dyn11 transformer produces a 30° displacement between HV and LV currents that must be removed before comparison.
• Zero-sequence current: an earthed star winding sources zero-sequence current for external earth faults which has no counterpart on the delta side.

Add CT ratio and phase errors, relay measurement error, and CT transient saturation during heavy through faults, and you have the reason the characteristic is biased (percentage restrained) rather than a simple threshold.

The Biased Characteristic

Slope 1 covers the steady-state error budget CT composite error, OLTC range, relay tolerance with margin. Slope 2, applied above a knee somewhere around 2–4 I_n, exists almost entirely for one reason: transient CT saturation during heavy external faults, where one CT saturates before the other and produces a large spurious differential current at high bias. Above that, an unrestrained (high-set) element, typically 8–10 I_n, trips without harmonic blocking on the assumption that no inrush or saturation condition can plausibly reach that magnitude.

Magnetising Inrush and Second-harmonic Restraint

When a transformer is energised, residual core flux and the point on wave at closing can drive the core deep into saturation, drawing a magnetising current that appears entirely on the energised side. To the differential relay this is indistinguishable from an internal fault in magnitude peaks of 8–12 times rated current are ordinary. It decays over seconds, not cycles.

The classical discriminant is harmonic content: inrush waveforms are heavily distorted and rich in second harmonic, while internal fault currents are not. A restraint threshold around 15% second harmonic relative to fundamental is typical. Modern low-loss core steels have pushed inrush second-harmonic content lower values below 10% are reported which is why relays offer cross-blocking between phases, and why waveform-based (gap detection or dwell-time) methods have become common alternatives. Recovery inrush after external fault clearance and sympathetic inrush from a parallel energised unit are the two conditions that catch out settings tuned only for initial energisation.

Vector Group Compensation and Zero-sequence Elimination

In electromechanical schemes, phase compensation was done by wiring the CTs: delta-connected CTs on the star winding, star-connected CTs on the delta winding. That connection did two jobs at once it corrected the 30° shift and it trapped zero-sequence current. Numerical relays do both in software, and this is where the trap lies: correcting the vector group does not automatically remove zero-sequence current. On a star winding with an earthed neutral, zero-sequence elimination has to be explicitly enabled, or an external earth fault on that side will produce differential current the relay has no reason to restrain.

Entering the wrong clock number is equally punishing and much easier to spot. The phase error produces a standing differential current proportional to load. So it shows up during load-current stability checks rather than waiting for a fault.

CT Ratio Correction and Amplitude Matching

A 63 MVA 132/33 kV transformer draws roughly 276 A on the HV side and 1102 A on the LV side. With, say, 300/1 and 1200/1 CTs, the secondary currents at rated load are 0.92 A and 0.92 A conveniently matched here, but in practice the available CT ratios are chosen for metering and switchgear reasons. The relay applies an amplitude matching factor to each winding to bring both to a common reference. Every relay imposes a limit on how far that factor can stretch (commonly around 0.25–4), and exceeding it is a design problem, not a settings problem.

5. Restricted Earth Fault Protection (64REF)

For an earth fault on an impedance or solidly-earthed star winding, the driving voltage falls linearly as the fault point moves toward the neutral. The current seen on the delta side falls faster still. The practical consequence is that a differential relay loses sensitivity over roughly the bottom 20–30% of the winding and that is precisely the region where insulation stress and mechanical clamping problems tend to show up.

REF closes that gap by comparing the residual current of the three phase CTs on the star winding with the current in the neutral CT. Because it is a balanced, single-winding scheme with no ratio, vector group or inrush complications, it can be set far more sensitively than the differential primary operating currents of 10–20% of winding rating are routine.

The high-impedance calculation is short but unforgiving. The stability voltage is the product of the maximum through-fault secondary current and the loop resistance of the CT secondary winding plus lead burden, with the setting voltage placed above it. The CT knee point placed at least twice the setting voltage.

6. Overfluxing / Overexcitation (24)

Core flux density is proportional to the ratio of applied voltage to frequency. Raise of the voltage or drop of the frequency saturates the core. Once saturated, flux spills out of the intended path into tank walls, core clamps and structural steel. Where it drives eddy currents and localised heating. Damage develops in minutes to tens of minutes. Which is far too slow for any current-based protection to notice, and far too fast to ignore.

The classic exposure is a generator step-up transformer during run-up, run-down, or load rejection, where the machine can be at reduced frequency or elevated terminal voltage while still connected. Transmission transformers are much less exposed, which is why 24 is not universally applied.

Transformer standards allow continuous operation at about 1.05 per-unit V/Hz at rated load and 1.10 per-unit at no load. Protection typically consists of a definite-time alarm just above the continuous limit. Also set an inverse-time V/Hz characteristic matched to the manufacturer’s overexcitation withstand curve, with an instantaneous element for severe excursions.

7. Backup Overcurrent, Earth Fault and Thermal Overload

Backup overcurrent on the HV side has to satisfy three constraints simultaneously, and plotting them on one time–current diagram is the only reliable way to see whether a setting works:

1. Above the transformer inrush point conventionally plotted at 8–12 times rated current for 0.1 s.
2. Above maximum permitted overload, including emergency ratings with all cooling in service.
3. Below the through-fault withstand curve from IEEE C57.109, which sets how long the transformer can survive a given external fault current without cumulative mechanical damage.

A standby earth fault relay (51G) on the neutral CT gives sensitive backup for earth faults anywhere on the earthed winding or its connected system. This is usually the last line of defence if REF is out of service.

Thermal overload (49) models winding hot-spot temperature from load current using a thermal replica, generally in line with the loading guidance of IEC 60076-7. It is normally arranged as a two-stage alarm then trip. It coordinates with the winding temperature indicator, which measures the same quantity by a completely different route.

8. Mechanical and Non-electrical Protection

It is worth stating plainly, because it surprises people new to transformer protection: on a large proportion of real transformer failures, the mechanical protection operates before the electrical protection does. An interturn fault involving a handful of turns produces a differential current that may sit comfortably below the pickup. But it produces gas immediately.

Gas collected at a Buchholz alarm should always be sampled rather than simply released. Its volume, combustibility and composition, taken with a dissolved gas analysis of the oil, distinguishes trapped air after a filtration or top-up from a genuine developing fault.

9. CT Requirements for Transformer Protection

Every function above depends on current transformers behaving predictably under conditions far from rated. Three requirements dominate:

• Transient performance for 87T. The differential relay must remain stable for the maximum external fault. That means the CTs on both sides must avoid saturating differently. In practice this is expressed through the accuracy limit factor and the connected burden. This is why a differential scheme sharing CT cores with a high-burden metering circuit is a bad idea.
• Matched cores for high-impedance REF. Identical ratio, identical class, low and known secondary winding resistance, and no other function sharing the core. Class PX/X exists precisely for this duty.
• Ratio consistency across the zone. Dual-ratio CTs with tap links are a recurring commissioning hazard the link position on site and the ratio entered in the relay must agree, and only a primary injection test proves it.

10. Tripping Philosophy, Lockout and Alarm/Trip Matrix

Almost all transformer protection trips are routed through a lockout relay (86T), for one reason: a transformer trip should never be reset and re-energised by a control action alone. The lockout forces a deliberate human decision, and it trips every breaker bounding the zone HV, LV, and any tertiary rather than only the one nearest the relay.

A transformer fault is never transient. Auto-reclose onto a faulted transformer converts a repairable fault into a replacement, so every transformer zone trip must block reclosing on the bounding breakers.

11. Commissioning Checks That Catch Real Errors

These are the checks that most often find something wrong, in roughly the order they should be done:

1. Primary injection through each winding to prove CT ratio, polarity and the wiring route from the CT terminal box to the relay. Secondary injection proves the relay; only primary injection proves the installation.
2. Verify the relay’s transformer data rated MVA, both winding voltages, both CT ratios, vector group / clock number, star-point earthing status against the nameplate, not against the design drawing.
3. Confirm zero-sequence elimination is enabled on every star winding with an earthed neutral.
4. Stability check on load. With the transformer energised and carrying load, read the differential and bias currents per phase from the relay. Differential current should be a small fraction of bias current in all three phases. A significant, load-proportional standing differential in all three phases points at vector group or matching factor; in one phase, at a CT connection.
5. Through-fault stability where a staged test is practical, or at minimum a review of the bias setting against the calculated maximum through-fault current and the CT saturation margin.
6. Energisation test. Close onto the unloaded transformer several times and capture the disturbance record. This is the only realistic proof that the harmonic restraint settings hold, and the records are worth keeping as a baseline.
7. Functional test of every mechanical contact to the lockout and to the SCADA point, individually, end to end.
8. Lockout test confirm 86T trips every bounding breaker and blocks reclose.

12. ANSI Device Number and Standards Reference

13. Frequently Asked Questions

Settings and thresholds quoted in this article are typical industry ranges given for illustration. Every setting must be derived from the specific transformer nameplate, CT data, system fault levels and the manufacturer’s withstand curves for the installation concerned.