SFRA Testing: A Practical Guide to Sweep Frequency Response Analysis

A transformer winding that has shifted 2 mm inside the tank will pass every conventional electrical test. The turns ratio is unchanged. The winding resistance is unchanged. The insulation resistance is unchanged. The DGA is clean if the movement hasn’t caused arcing yet. Energize the transformer, run it for years, and the shift slowly progresses — until one day a through-fault on the system delivers the force that finishes the job, and the transformer fails catastrophically.

SFRA is the test that catches the shift while it’s still 2 mm.

Sweep Frequency Response Analysis works by treating the transformer as a complex network of resistances, inductances, and capacitances and probing that network across a wide frequency range. Mechanical movement inside the transformer — winding displacement, core shifts, loose clamping — changes the network’s R, L, and C values, which changes its frequency response, which shows up as a difference between today’s measurement and a baseline taken when the transformer was known-good.

This is a guide to running and reading SFRA properly. It assumes familiarity with transformer testing in general — the value here is in the interpretive framework: connecting what you see on the screen to what’s happening inside the tank.

What SFRA Is Actually Doing

A transformer winding is electrically more than just resistance and inductance. Between every pair of turns there’s capacitance. Between every turn and the core, capacitance. Between primary and secondary windings, capacitance. The result is a complex RLC network whose behavior changes with frequency in characteristic ways.

The SFRA instrument injects a low-voltage sinusoidal signal (below 10 V) into one terminal of the transformer and measures the response at another terminal. It sweeps the frequency from around 20 Hz up to typically 2 MHz, recording the transfer function — the magnitude (in dB) and phase relationship between input and output — at thousands of frequency points along the way.

The resulting trace is a fingerprint of the transformer’s mechanical and electrical geometry. Two transformers built identically will produce very similar traces. The same transformer, tested today and a year ago, should produce nearly identical traces if nothing has changed mechanically. When traces differ — between a baseline and a current measurement, between sister units, or between phases of the same transformer — the difference points to a change in the underlying R, L, or C values.

The key insight is that mechanical changes drive the electrical changes. A winding that has bowed outward from a short-circuit force changes the capacitance between turns. A core leg that has shifted changes the magnetic coupling. A loose clamping that allows a winding to settle changes both. SFRA detects mechanical movement indirectly by detecting the electrical changes it causes.

Sensitivity is high: displacements of 1-2 mm in winding position can produce measurable frequency response shifts. That’s well below the threshold at which conventional tests would notice anything wrong.

What SFRA Catches That Other Tests Miss

Most transformer tests are insensitive to mechanical condition. TTR catches turns problems and connection errors. Winding resistance catches loose connections and broken conductors. Insulation resistance and power factor catch insulation degradation. Excitation current catches gross core problems. DGA catches active fault gases. None of these catches a winding that has moved.

Winding movement matters for two reasons. First, a deformed winding has reduced short-circuit withstand capability. The next through-fault that hits the transformer may push the displaced winding past its mechanical limit and cause failure. Second, severe deformation can produce internal contact between conductors or insulation damage that triggers a dielectric failure in service.

The events that cause winding movement:

Through-fault currents. Short-circuit forces inside the transformer are proportional to current squared. A high-current external fault delivers enormous mechanical impulse to the windings. The IEEE guide recommends SFRA after any through-fault exceeding 70% of rated short-circuit withstand current.
Transport damage. Road or rail movement, especially with shock during loading, can shift windings. SFRA before and after transport (or before and after a relocation) is standard practice on large units.
Through-fault history followed by routine operation. A winding stressed but not failed by an earlier fault may settle gradually under normal load until the next fault completes the failure.
Loss of clamping. Pressing systems can loosen over decades of thermal cycling. A winding that was tight in 1985 may not be tight today.
Earthquakes or other seismic events. Less common but worth a check after.

Other triggers that warrant SFRA: Buchholz relay operation (especially gas accumulation without obvious cause), sudden pressure relief operation, unexplained DGA gas increases (particularly acetylene, which suggests arcing), audible noise changes that might indicate loose core or winding components.

The Standards That Govern SFRA

Three documents control the practice:

IEC 60076-18 — Measurement of frequency response. Originally 2012, updated 2025. The international standard for how to perform the measurement: connection methods, frequency ranges, voltage levels, documentation requirements.

IEEE C57.149-2012 — Guide for the Application and Interpretation of Frequency Response Analysis for Oil-Immersed Transformers. The US standard. Detailed coverage of interpretation, test configurations, and case studies. Updated periodically; check for current revision.

CIGRE Technical Brochures 342 (2008) and 445 (2010) — Comprehensive practical guidance on interpretation, frequency band assignment, and case studies. Not standards per se, but widely used as reference.

DL/T 911-2004 — Chinese electric power industry standard with specific deviation criteria for winding deformation assessment.

The standards agree on the measurement technique. They differ slightly on the interpretation specifics — frequency band boundaries vary by a few hundred Hz, dB deviation thresholds vary by a couple of dB. In practice, all four are workable; pick one as primary reference and stay consistent.

Test Configurations: What to Connect to What

A complete SFRA test on a three-phase transformer is multiple measurements with different terminal connections. IEEE C57.149 specifies nine standard tests for a Dyn1 configuration, but the practical core comes down to three test types:

Open-circuit (end-to-end) tests. Measure the frequency response across one winding (terminal to neutral, or terminal to terminal) with all other windings open-circuited. Done separately for each phase of the HV winding, then each phase of the LV winding. Six tests total on a three-phase transformer with neutral brought out. Sensitive to core condition and to overall winding integrity.

Short-circuit (HV-LV) tests. Measure across the HV winding with the corresponding LV phase shorted. Done for each phase. Three tests. The short circuit eliminates the core’s contribution from the response, leaving the winding’s own electrical signature exposed. This is the most sensitive test for winding deformation — the LV short means the response reflects only the HV winding’s geometry and the leakage path between HV and LV.

Capacitive inter-winding tests. Less common in routine use, sometimes specified for specific diagnostics. Measure the capacitive coupling between HV and LV windings.

The short-circuit tests are particularly important for detecting LV winding deformation, which is the most common winding-movement failure mode in three-phase core-form transformers stressed by through-faults.

A few practical setup points that affect results materially:

Coaxial cables, matched length. The instrument’s signal and measurement cables are typically 50-ohm coax. Use the manufacturer’s cables. Don’t substitute. The cable characteristic impedance and propagation delay are part of the calibration.

Grounding. Single-point grounding back to the instrument. Multiple ground paths create ground loops that show up as response artifacts at specific frequencies.

Tank earthing. The transformer tank should be solidly earthed to the instrument’s reference ground.

Tap position. Record and report the tap position for every measurement. SFRA traces shift with tap position because the effective winding changes. Comparing traces taken on different taps gives misleading differences. Standardize on a consistent tap position (typically nominal/principal) for baseline and trending.

Oil condition. SFRA traces differ slightly between oil-filled and oil-empty conditions because the dielectric environment changes. Don’t compare an oil-filled measurement to an oil-empty one unless you specifically want to characterize that difference.

Bushing accessories. Cables, switchgear, and other terminations affect the response, particularly at higher frequencies. Disconnect everything beyond the transformer terminals. Document what’s connected.

The Frequency Bands and What Lives in Each

This is the heart of interpretation. The frequency spectrum maps onto physical components of the transformer. A deviation in a specific frequency band points to a problem in the specific component that dominates that band.

Below ~2 kHz (low frequency band). The core dominates. Magnetizing inductance is the largest impedance at low frequencies. Deviations in this range indicate:

Core deformation or movement
Open circuits in the magnetic path
Shorted turns (which short out part of the magnetizing inductance)
Residual magnetism affecting the response

A typical low-frequency response has a decreasing magnitude as frequency rises, reaching a minimum at the resonance between the magnetizing inductance and bulk transformer capacitance. The shape of that minimum and its frequency location are characteristic of the core.

A normal but counterintuitive feature: on a three-limb core, the center phase (B) often has a noticeably different low-frequency response than the outer phases (A and C). The B phase core leg has symmetric flux return paths through both outer legs; the A and C phases have asymmetric return paths. This produces a different resonance structure for the center phase. A test engineer who sees the B phase trace and thinks “that doesn’t match A and C” may be looking at completely normal three-limb behavior, not a fault. The comparison that matters is B today vs B baseline, not B vs A.

~2 kHz to ~20 kHz (low-to-mid frequency band). The bulk winding and the interaction between core and winding dominate. Shunt impedance effects, the resonance between winding inductance and inter-winding capacitance. Deviations here indicate:

Bulk winding movement
Inter-winding (HV-LV) displacement
Loss of clamping on a bulk scale

~20 kHz to ~400 kHz (mid-frequency band). The main winding structure dominates. Inter-disc capacitance, individual turn-to-turn relationships. This is the most sensitive band for winding deformation. Deviations here indicate:

Radial deformation (winding bowed outward or inward)
Axial deformation (winding shifted along its axis)
Localized disc displacement
Loss of clamping causing winding settling

~400 kHz to ~2 MHz (high frequency band). Internal winding connection effects, lead arrangements, and the finest-scale geometric details. Deviations here often indicate:

Tap lead movement
Internal connection issues
Local winding defects too small to show in lower bands

These boundaries are approximate. Different standards draw them slightly differently. The IEEE band boundaries are the most commonly cited; CIGRE breaks them somewhat differently. In practice, what matters is the consistency within a single program — if you’re comparing today’s traces to last year’s, the band definitions are arbitrary as long as they’re consistent.

Note that for windings rated less than 72 kV, IEC recommends extending the sweep to 2 MHz. The high-frequency content carries more diagnostic information on smaller, lower-voltage units because their winding geometry has finer features that respond at higher frequencies.

Comparison Methods: Three Ways to Know What Normal Looks Like

SFRA is fundamentally a comparative test. A single trace, taken in isolation, tells you almost nothing. You need a reference to compare against. There are three approaches:

Time-based comparison. Today’s measurement against a baseline taken when the transformer was known-good. This is the gold standard for in-service condition monitoring. The baseline is typically taken at commissioning, before the transformer enters service. Subsequent tests are compared against this baseline at standard intervals or after triggering events. Time-based comparison requires that the baseline was taken with comparable equipment and methods — a baseline taken with a different test set, different cables, or different tap position is contaminated and reduces sensitivity to real changes.

Sister-unit comparison. Today’s measurement against a measurement on an identical (or nominally identical) sister transformer. Useful when no baseline exists — for example, on an older transformer where SFRA was never run, or after a transformer has been moved without a fresh baseline. Sister-unit comparison is less reliable than time-based because no two transformers are perfectly identical, but it can detect gross differences.

Phase-to-phase comparison. Today’s three phases compared against each other on the same transformer. Useful when no baseline and no sister exist. The limitation is the natural difference between phases (especially the B phase center-leg behavior described above) — phase-to-phase comparison is most reliable for detecting differences that are sharp and localized to one phase, not for general assessment.

The strongest interpretation combines all three. A deviation that shows up in time-based, sister-unit, AND phase-to-phase comparison is almost certainly real. A deviation that appears in only one comparison method may be an artifact.

Reading Differences: How Much Deviation Matters

The interpretive question is “how much difference between traces is meaningful?” Standards and practitioners converge on roughly this framework:

Below 3 dB deviation. Generally within measurement repeatability. Probably not significant. Could be due to small differences in setup, slight temperature differences, instrument noise. Don’t act on changes below this threshold.

3 to 6 dB deviation. Investigate further. Possibly significant. Look at whether the deviation is in a specific frequency band (pointing at a specific component), whether it correlates with any other diagnostic data, whether the transformer has been through any events that could cause a real change. May indicate developing issues.

Above 6 dB deviation. Strongly suggests real mechanical change. Action warranted — at minimum, increase monitoring frequency. May indicate movement requiring intervention.

These thresholds are practical guidelines, not absolutes. Transformer criticality affects acceptable risk: a 5 dB deviation on a major transmission autotransformer justifies immediate investigation, while the same deviation on a small distribution unit may permit continued monitoring with shorter inspection intervals.

The IEC 60076-18 approach uses correlation coefficient calculation between reference and measured traces, with thresholds for the correlation value in different frequency bands. This is more rigorous than visual comparison and is built into modern SFRA software. The dB-deviation approach and the correlation approach generally agree on which deviations are significant; the correlation method is more quantitative for reporting.

Where the deviation lies in the spectrum tells you what kind of fault to suspect, using the frequency band map above. A 6 dB deviation in the 50-200 kHz range with normal low-frequency response points at winding deformation, not core movement. A deviation across the entire low-frequency range with normal mid- and high-frequency response points at core problems.

Temperature and Other Practical Variables

A few real-world factors that affect SFRA results:

Temperature. Less of a factor than for resistance-based tests. The IEEE guide notes that large temperature differences (more than about 10°C) between two measurements may slightly affect the high-frequency response. For practical purposes, ignore temperature unless the difference between today’s measurement and the baseline is dramatic. Record the temperature; don’t worry about correcting for it.

Residual magnetism. Strong residual core magnetism affects the low-frequency response. After a winding resistance test (which leaves the core heavily magnetized), demagnetize before running SFRA. This is the same demag step the TTR test needs — sequence your tests so it gets done once and serves both.

Test repetition. Run each measurement twice. If the two repetitions don’t overlay almost exactly, something in the setup is unstable — bad connection, cable issue, noisy environment. Don’t move on until repetitions are consistent.

Documentation. Photograph the test setup. Record temperatures, tap position, oil status, what’s connected to bushings, weather if it’s a field test. The next person running SFRA on this transformer — in five years, when conditions of the original test are forgotten — will thank you. Or curse you.

The Diagnostic Workflow

Pulling it together, a real SFRA program looks like this:

Establish a baseline at commissioning or first opportunity. All standard test configurations, careful setup, full documentation. This baseline is the reference forever after.
Periodic measurements at intervals appropriate to transformer criticality. Major transmission units every 3-5 years; distribution units less frequently.
Triggered measurements after any event that could cause mechanical change: through-faults above 70% of rated withstand, Buchholz operation, pressure relief, transportation, suspect DGA changes, audible anomalies.
Compare today’s measurement to baseline. Look for deviations by frequency band. Apply the dB or correlation threshold framework.
Cross-correlate with other tests. SFRA findings are stronger when supported by other data. A deviation in the winding deformation band, combined with elevated DGA acetylene, combined with elevated leakage reactance, paints a clear picture. SFRA in isolation can flag a suspicion; multiple tests together confirm it.
Decide. Continue normal monitoring, increase monitoring frequency, schedule internal inspection, or take the transformer out of service. The SFRA result is one input to this decision, not the whole answer.

The Takeaway

SFRA is the test that catches mechanical condition — the dimension of transformer health that other tests can’t see. Done well, it detects winding movement, core shifts, and clamping issues at displacement scales that conventional tests miss entirely. Done badly, it produces traces that vary from one measurement to the next for setup reasons and obscure real changes in noise.

The discipline that separates the two is concrete: establish baselines on new transformers, hold setup conditions constant across measurements, compare against baseline and sister units and phases together, read deviations by frequency band against a consistent threshold, and combine SFRA findings with other diagnostic data rather than treating them in isolation.

The cost of an SFRA test is low. The cost of a transformer that fails in service because winding movement went undetected for years is enormous. The economics push hard toward running SFRA more often, on more transformers, with more discipline — particularly after the events that cause real mechanical change. The instrument is sensitive enough; the practice has to match.