SFRA and the Operator's Diagnostic Gap

Introduction

Transformer fleet operators rely on a set of established diagnostic tools to assess mechanical and electrical health: dissolved gas analysis, insulation resistance and power factor testing, and sweep frequency response analysis, among others. Each test targets a specific failure mode and each has a specific operating constraint. This paper focuses on SFRA, describes what it measures and when it can be measured, and shows why continuous multi-signal monitoring addresses a different point in the same failure sequence.

The argument advanced here is not that SFRA is limited in accuracy. It is not. The argument is that SFRA and continuous monitoring observe different physical quantities at different stages of mechanical degradation, and that the stage an operator can act on economically is the earlier one. A secondary argument follows from the first. SFRA is an event-driven test, and an operator managing a live fleet currently has no dependable way to generate the event.

SFRA: Technical Overview

SFRA measures the transfer function of a transformer winding across a swept frequency range, typically from 20 Hz to 2 MHz. A low-voltage signal is injected at one terminal and the response is measured at another, producing a signature curve determined by the winding's distributed resistance, inductance, and capacitance network. Because that network is a direct function of physical geometry, turn spacing, and dielectric properties, any change in winding shape or position alters the signature.

Different frequency bands carry different diagnostic sensitivity, summarized below.

SFRA is governed by IEEE C57.149-2024 and IEC 60076-18, with foundational work established by CIGRE working group A2.26. It is well validated against known mechanical fault types, including winding buckling, axial displacement, core displacement, and shorted or open turns. Its diagnostic depth across those fault types exceeds that of any continuous method available today, and this paper takes that as established.

The Comparative Nature of SFRA Results

SFRA has no absolute fault threshold. A signature curve, on its own, does not indicate whether a winding is healthy or damaged. Diagnostic conclusions require one of three comparisons.

• Time-based comparison, against a signature taken when the unit was known to be healthy.
• Type-based comparison, against a sister or identical unit.
• Phase-based comparison, across the three phases of the same transformer.

Each approach depends on the availability and quality of a comparison point. A unit commissioned without a baseline signature, or a unit with no true sister unit in the fleet, is harder to interpret with confidence. Even with a valid comparison, interpretation requires expert judgment, and a deviation in the curve does not always map cleanly to a specific fault location or severity. This is a known and accepted characteristic of the test, addressed directly in the governing standards, and it is not offered here as a criticism.

When SFRA Can Be Run

SFRA is an offline test by design. The transformer must be de-energized, isolated, and made safe for low-voltage signal injection. That requirement is what makes the measurement controlled and repeatable, and the precision of the result depends on it. It also restricts testing to occasions when the unit is out of service.

• Commissioning, to establish a baseline signature.
• After a suspected through-fault event, such as a nearby short circuit.
• After transport or relocation, when mechanical shock is a concern.
• During a scheduled maintenance outage.

Practice varies. Some operators test critical units on a defined outage cycle. Others test only by exception. The specific number of tests per unit is not the point of this paper, and any figure quoted as typical would be wrong for some fleet. The point is the unit of measure. Whatever the interval, it is measured in years, because it is bounded by outage scheduling rather than by engineering preference.

Throughout this paper, coverage refers to time. SFRA's diagnostic depth is not at issue anywhere in these pages, and where depth is discussed it is named as depth. Coverage means how much of a transformer's operating life is under observation for mechanical condition. An offline test, however precise, provides intermittent coverage by construction. Section 5 explains why an interval measured in years is the wrong scale for the mechanism this paper is concerned with.

The Precursor and the Result

Mechanical failure of a transformer winding follows a sequence. Setting that sequence out explicitly is what separates this argument from a comparison of measurement resolution.

A winding is held in compression. Axial clamping force preloads the winding stack against the movement that electromagnetic forces would otherwise produce. That preload does not hold constant across an operating life. Cellulose insulation and pressboard shrink through drying and thermal cycling, and the clamping force relaxes as they do. The winding at this stage has not moved. Its geometry is unchanged. What has changed is how much freedom it has to move when force is applied.

Force is applied by through-fault current. Radial and axial forces on a winding scale with the square of current, so a fault current several times rated current produces forces more than an order of magnitude above normal duty. A fully clamped winding absorbs that event. An under-clamped winding may not, and the displacement, tilting, or buckling that results is the geometric change SFRA is built to detect.

The sequence therefore has a precursor stage and a result stage. SFRA measures geometry, which means it observes the result. It is definitive at that stage. It is silent at the stage before it, and this is not a shortcoming of the instrument. A winding that has lost half its clamping preload but has not yet moved presents very nearly the same resistance, inductance, and capacitance network it presented when new. There is little for the transfer function to reveal, because nothing geometric has happened yet.

Continuous vibration monitoring observes the precursor stage, because clamping force and mechanical stiffness are the same physical property viewed from two directions. The winding and core assembly is a mass and stiffness system driven continuously by electromagnetic forces whenever the unit is energized. Force on the winding follows the square of load current and therefore acts at twice line frequency. Core magnetostriction acts at twice line frequency and its integer harmonics. In a healthy unit, the excitation is deterministic, phase-locked to the supply, and confined to integer orders of twice line frequency.

That is what makes the response diagnostic. Three distinct observables change as clamping preload falls, and they change for three different reasons.

Amplitude. Lower stiffness produces greater displacement for the same applied force, and it lowers the resonant frequencies of the assembly. This is the most direct consequence of reduced preload and the easiest to measure. It is also the most confounded, because amplitude rises with load current independently of any change in mechanical condition, which is why amplitude alone is a weak indicator and why load and thermal context are required to interpret it.

Harmonic content. As preload falls, contact between pressboard, spacers, and conductor discs stiffens and softens through the cycle rather than behaving linearly. A nonlinear stiffness distorts the response waveform and redistributes energy into higher integer orders of the driving frequency. The energy remains phase-locked to the supply, but its distribution across orders changes. The ratio between orders is far less sensitive to load level than absolute amplitude, so it survives normalization.

Non-harmonic content. Once preload falls far enough that contact is intermittently lost through part of the cycle, the structure begins generating impact and friction events that are not phase-locked to the supply at all. This energy appears between the integer orders as broadband and non-integer content. Its diagnostic value comes from the fact that the excitation cannot produce it. The electromagnetic and magnetostrictive forces acting on an energized transformer are strictly integer-ordered, so off-order energy was created by the structure rather than delivered to it. There is no benign explanation from load.

The three observables also form an escalation. Amplitude moves first and ambiguously. Harmonic redistribution follows as contact stiffness becomes nonlinear. Non-harmonic content appears when contact is being lost outright, which is the state immediately preceding the freedom to move that a through-fault converts into displacement. Read together, and read against known load, ambient temperature, and magnetic signature, they describe not only that the mechanical state has changed but how far the change has progressed. The whole of this occurs while the geometry, and therefore the SFRA signature, remains unchanged.

This is the substantive claim of this paper, and it is worth stating in plain terms. Continuous monitoring is not a lower-resolution SFRA. It does not measure the transfer function, it does not localize a deformation to a winding disc, and it should not be described as approximating a result that it does not compute. It measures a different physical quantity, earlier in the same failure sequence, at a cadence an offline test cannot reach.

Read down that table and the division of labor is clear. Stages 4 and 5 belong to established, standards-backed tests, and those tests address them well. Stages 1 through 3 are the stages at which intervention is cheapest and the asset is still fully recoverable, and in most fleets they have no continuous observer at all. That is the gap this paper is about.

The Trigger Problem

The second argument is operational rather than physical, and in practice it is the one that determines whether a mechanical condition is found at all.

SFRA is invoked by cause. Commissioning and post-transport testing are scheduled by event. Post-fault testing requires that someone knows a fault occurred and judged it severe enough to justify an outage. Outage-cycle testing requires an outage. In every case other than commissioning, something must first direct attention to the unit.

For the mechanical degradation path described in Section 5, that something is usually absent. Three reasons compound.

Through-fault duty is cumulative and rarely attributed per unit. Protection systems record operations, but few fleets maintain a per-transformer ledger of fault current magnitude and duration over an operating life. Mechanical damage from through-faults accumulates. A winding is rarely deformed by one event. It is loosened by many and then displaced by one that would have been survivable earlier. Without a per-unit record, the operator has no basis for identifying which units have absorbed the most duty.

Dissolved gas analysis does not observe this failure mode early. DGA is an effective and widely deployed test, and it detects thermal and electrical faults through the gases their byproducts generate. A winding that has lost clamping pressure but has not arced and is not locally overheating generates no characteristic gas signature. DGA becomes informative at stage 5 of the sequence, once insulation damage at a displaced region begins producing detectable byproducts. By then the geometric change has already occurred.

No one runs an offline test on a unit nobody suspects. Outage time is expensive and crews are constrained. A unit with clean oil results, no recorded fault history, and no operational complaint does not get pulled. This is a rational allocation of limited resources, and it is exactly why an under-clamped winding can sit in service, undetected, until a through-fault converts the condition into a deformation.

The consequence is a specific and correctable failure of process. The most valuable moment to run an SFRA test is on a unit that has lost clamping pressure and has not yet been deformed, because at that moment the finding is actionable and the repair is a reclamp rather than a rewind. That is also precisely the moment at which no operator has any reason to schedule the test.

Continuous monitoring resolves this by generating the cause. When a unit's vibration response shifts in a manner consistent with reduced stiffness, and the shift persists and progresses against known load and ambient conditions, the operator has a defensible, unit-specific reason to schedule an outage and call for an SFRA test. The offline test then does what it does better than anything else, which is to confirm and localize with precision, on a unit selected by evidence rather than by calendar.

Positioned this way, continuous monitoring does not compete with SFRA for the same budget line or the same decision. It converts SFRA from a test that is scheduled or reactive into a test that is targeted. Operators are not asked to give up anything. They are asked to stop selecting units at random.

Continuous Multi-Signal Monitoring

Continuous condition monitoring observes the transformer at all times, under real operating conditions, rather than at scheduled snapshots. A four-signal approach, fusing triaxial vibration, thermal, magnetic field, and local weather data, provides complementary views of the transformer's condition.

Triaxial vibration. Captures the mechanical response of the winding and core structure to load-driven electromagnetic forces and core magnetostriction, both of which act at integer orders of twice line frequency. Amplitude, harmonic distribution, and non-harmonic content are evaluated separately, because each responds to a different stage of clamping loss and each carries a different sensitivity to load, as described in Section 5.

Thermal monitoring. Tracks temperature distribution and rate of change, providing context for load-dependent behavior and identifying localized heating that can accompany connection degradation or insulation issues.

Magnetic field sensing. Detects anomalies associated with circulating currents, core grounding faults, and stray flux conditions that are not always visible to vibration or thermal signals alone, and provides an independent observation of loading.

Local weather data. Provides the environmental context, ambient temperature, and loading conditions needed to distinguish a genuine condition change from a normal, expected response to load or weather variation.

Fusing the four matters because a single signal is ambiguous. A rise in vibration amplitude that tracks a rise in load is expected behavior. The same rise at constant load, constant ambient temperature, and constant magnetic signature is not. Separating the two requires the context signals, and this is the reason a vibration-only approach produces alerts an operator learns to ignore.

Because these signals are captured continuously, the system characterizes each individual unit under its own real operating conditions rather than relying on a single offline measurement taken years earlier. Deviations from that characterization, and the rate at which they develop, become the basis for anomaly detection and trending.

What Continuous Monitoring Does Not Do

A precise scope is necessary for this argument to hold up in front of a test engineer.

Continuous multi-signal monitoring does not replicate SFRA's diagnostic resolution and does not attempt to. Once a deformation has occurred, SFRA localizes the affected region of a winding by frequency band with a geometric specificity that no continuous surface measurement produces. SFRA also separates fault types that continuous monitoring does not resolve individually: axial displacement, radial buckling, core displacement, shorted turns, open circuits, and lead or tap changer problems each present differently in an SFRA signature. A vibration-derived stiffness change indicates that the mechanical state has changed. It does not, on its own, tell a technician which of those conditions is present.

SFRA is also independent of operating conditions. It is a controlled injection test that does not require load to be present, does not require a monitoring history, and provides a self-contained phase-to-phase comparison. Continuous monitoring requires time to establish a per-unit reference and requires load and weather context to interpret.

When a unit is open and a technician is about to reclamp a winding, disc-level resolution is exactly the right input, and nothing in this paper substitutes for it. The question is what an operator does in the years when no one is opening the unit, and which unit should be opened first.

The Dynamic View: Load, Weather, and Operating Condition

A further structural difference follows from the offline requirement. SFRA is performed at zero load, in a controlled state chosen for repeatability, and that control is what makes its results trustworthy. It is not designed to show how a winding responds as load current rises through a summer peak, as ambient temperature swings between a Mongolia winter and an Arizona summer, or as thermal cycling accumulates across thousands of load cycles.

This matters because a mechanical condition can manifest differently, or only become apparent, under specific load or thermal conditions. A loose winding may show a vibration signature change that is negligible at light load and pronounced at peak load, because the electromagnetic forces driving winding motion scale with the square of current. A connection degrading from thermal cycling may show a temperature rise that appears only during rapid load swings. None of this is visible to a test performed once, at zero load, because that is not the question the test was built to answer.

The comparison is summarized below.

Specificity and Decision Utility

Diagnostic resolution and decision utility are two different axes, and separating them explicitly clarifies where each method earns its place.

SFRA answers a geometric question with real precision: has this winding's shape changed, and if so, in which region. The questions an operator carries between outages are different. Is this condition stable or getting worse. How fast is it changing. Does it warrant action now or observation for another season. Which of several hundred units should receive attention first. Those are rate and severity questions, and one measurement, however precise, does not establish a rate.

A confirmed deformation, precisely located, still leaves the operator without the information needed to prioritize it against every other unit in the fleet. A deformation that has been stable for ten years and a deformation that began last month can produce similar signature deviations while representing very different levels of urgency. Distinguishing them requires a second data point, and a second SFRA test is rarely available on a timescale short enough to establish a rate before a decision is due.

Disc-level location is decisive for the technician who will do the work. It maps less directly onto the actions available to an operator between outages, which are a small set: continue the unit in service and monitor, reduce loading, schedule an outage for inspection or repair, or plan replacement. Each of those decisions is driven by severity and trajectory rather than by which disc within the winding is affected.

This is why continuous monitoring's lack of geometric localization costs less than it first appears. The system trades geometric precision for a rate. Because vibration, thermal, and magnetic field signals are captured continuously against known load and weather conditions, the system observes not only that a condition exists but how it is behaving over time and whether it correlates with load, ambient temperature, or seasonal pattern. That rate is the input an operator needs to decide whether a unit can run through the next peak season or needs to move up the inspection queue now.

The comparison below sets the operator's questions against what each method is built to answer.

None of this diminishes SFRA's diagnostic accuracy. It identifies when that accuracy is decisive, which is once a unit has been selected and taken out of service. Continuous monitoring answers the earlier question, which is which unit to select.

Evidence and Validation

A claim about a failure sequence deserves a statement of the evidence behind each method, and the two evidence bases are not equivalent in kind.

SFRA carries an unusually strong validation record. IEEE C57.149-2024 and IEC 60076-18 define the measurement and its interpretation. CIGRE working group A2.26 established the foundational correlation work. Decades of field results have been checked against physical teardown, which is the strongest form of validation available for any transformer diagnostic. Any method proposed alongside SFRA should be measured against that standard rather than compared with it selectively.

Continuous vibration-based condition monitoring does not carry an equivalent standards history, and stating otherwise would not survive scrutiny. IEEE C57.143 provides guidance on the application and specification of monitoring equipment for liquid-immersed transformers, but no standard defines a mechanical pass or fail threshold from surface vibration, and none is likely to, because the response is unit-specific. The evidence base is therefore empirical and longitudinal rather than normative.

VIE's evidence base consists of the following.

• A monitored population of more than 1,000 transformers, spanning over 3 gigawatts under management, across utility, data center, oil and gas, and industrial customers, in service since commercial launch in 2020.
• Deployment across a wide environmental range, from Mongolia to Arizona, which is the condition set required to separate genuine mechanical change from ambient and load-driven response.
• Winding health metrics that have been correlated against independent laboratory measures of insulation condition, including insulation resistance, Furan content, and degree of polymerization.
• A physics-based rather than learned-baseline model, meaning the system computes expected mechanical response from first principles and does not absorb an existing fault into a normal condition, which is the principal failure mode of statistically trained monitoring systems commissioned on units already degraded.

The honest summary is this. SFRA's validation is deeper and older, and it is specific to geometry. Continuous monitoring's validation is longitudinal and specific to mechanical state under operating conditions, which is a body of evidence that only accumulates through fleet deployment over time and that no offline test can generate. The two are complementary because the underlying evidence is complementary.

Positioning by Role in the Transformer Lifecycle

The distinctions in this paper resolve into a straightforward positioning. The right diagnostic tool depends on which decision is being made, at which stage of the failure sequence, and how often that decision comes due.

For the transformer operator, the implication is direct. Continuous multi-signal monitoring is the ongoing source of mechanical condition information, because it observes the stage of the failure sequence at which an operator can still act cheaply, and because it is the only source available at the cadence fleet decisions require. SFRA remains the sharper instrument whenever a unit is opened, and it is the correct next step once continuous monitoring has identified a unit worth opening.

Conclusion

SFRA is an excellent, well-validated test with greater diagnostic depth than any continuous method available today. Its comparative structure and offline requirement are not flaws. They are the conditions that make the result precise. Nothing in this paper argues for running it less carefully or trusting it less.

What this paper argues is that SFRA measures geometry, and geometry changes late. Clamping pressure relaxes first, stiffness falls, and only then does a through-fault convert a loosened winding into a displaced one. Continuous vibration monitoring observes the earlier stage because stiffness governs the response of an energized assembly to forces that are present every hour of every day. This is a different observable, not a degraded version of the same one.

It also argues that SFRA is invoked by cause, and that operators currently lack a mechanism for producing the cause on the mechanical degradation path. Through-fault duty goes unattributed, dissolved gas analysis becomes informative only after insulation damage begins, and no one schedules an outage for a unit nobody suspects. The result is that the most valuable SFRA test, the one performed on an under-clamped winding before it deforms, is the one least likely to be ordered.

VIE Technologies closes that loop. Continuous, non-invasive monitoring across vibration, thermal, magnetic field, and weather signals identifies which unit has changed, how quickly, and under what conditions. That information supports the fleet decisions an operator makes every week. It also supplies the justification for an SFRA test on a specific unit at the moment the finding is still actionable and the repair is still a reclamp. SFRA stays where it is decisive. The years in between stop being unobserved.