Winding Health and the Diagnostic Evolution

How Winding Insulation Degrades Over Service Life

The operational lifetime of a power transformer is defined by the integrity of its winding insulation system. Structural components, tank assemblies, and core laminations exhibit multi-decade stability. Winding insulation does not. It undergoes continuous, irreversible chemical degradation from the moment the asset is energized.

Degradation is a chemical process, and its rate is governed by temperature. The Arrhenius relation makes the reaction rate exponential in temperature. It does not make aging exponential in elapsed time. IEEE C57.91 expresses the practical consequence for cellulose: the rate of insulation aging approximately doubles for every 6 °C to 8 °C of hot-spot temperature rise, depending on paper type and temperature range [3].

The observed service-life pattern follows from that rate law. Through the early service life, winding health stays broadly stable and the insulation experiences only minor thermal and mechanical conditioning. As cumulative thermal exposure accrues, oxidation, hydrolysis, and electrical transients progressively break the long polymer chains of the cellulose, and the mechanical strength of the paper falls.

In the terminal phase, the insulation has lost structural elasticity and tensile strength. At this stage a normal operational force or a minor external system fault can trigger local dielectric breakdown, turning a routine mechanical event into a non-recoverable phase-to-ground or turn-to-turn fault.

Rate is set by hot-spot temperature, not by load

Loading alone does not set the aging rate. Per IEEE C57.91 and IEC 60076-7, the winding hot-spot temperature is a sum: ambient temperature, plus top-oil rise over ambient, plus hot-spot rise over top oil [3][4]. Only the two rise terms are set by loading. Ambient enters degree for degree, which means the doubling rule that governs overload governs climate identically.

The arithmetic is not marginal. Two identically loaded units in yards whose summer ambient differs by 25 °C sit 25 °C apart at the hot spot. At a doubling interval of 6 °C to 8 °C, that is a difference in instantaneous aging rate of roughly nine to eighteen times. VIE monitors transformers from Mongolia to Arizona. At identical duty, those are not the same asset at the same age.

Temperature is not the only environmental term. IEC 60076-7 expresses the relative aging rate of cellulose as a function of temperature, moisture content, and oxygen content, not temperature alone [4]. The moisture dependence is steep. The rate of thermal deterioration of the paper is proportional to its water content, and mechanical life falls by half for each doubling of that content. Moisture ingress is largely a weather process. A free-breathing conservator inhales humid air on every cooling cycle, and the depth and frequency of those cycles is set by the diurnal and seasonal temperature swing at the site.

The mechanical vector is environmental as well. Thermal cycling amplitude works the clamping structure through repeated expansion and contraction. Storm activity governs through-fault frequency, and each through-fault delivers an electromagnetic shock to a winding stack that may already have lost clamping pressure. Lightning density, ice loading, and wind exposure are properties of the yard, not of the load profile.

These terms are not independent and they do not offset. Peak load and peak ambient coincide seasonally, so the worst thermal conditions of the year stack rather than cancel.

Two units of identical age can differ by many years of effective aging, set by accumulated hot-spot exposure, moisture history, and mechanical event count. Loading is one input. Site environment is another of comparable magnitude. Neither appears on the nameplate.

Why age is a weak predictor even though aging is real

Population statistics complicate any attempt to infer condition from age. In the same CIGRE A2.62 dataset, the hazard curve for all major failures was close to random, rising only slowly with age. Only failures severe enough to result in the unit being scrapped showed a clear age dependency, with the hazard rate accelerating after roughly twenty years of service [1].

Both findings are true at once. The degradation mechanism is real, irreversible, and well understood. Age is nonetheless a weak discriminator of which specific unit is at risk, because age does not carry the thermal history that actually drives the chemistry. That gap between a real mechanism and a weak population-level predictor is the entire case for measuring condition on each unit rather than inferring it from a nameplate date.

Winding Insulation Methods in Oil-Submerged Power Transformers

Oil-submerged power transformers use a dual-phase insulation system that combines solid dielectric structures with fluid-phase insulation.

Cellulose paper wrapping. High-purity, unbleached kraft paper tape is wound continuously around the copper or aluminum winding conductors. The paper acts as a mechanical carrier and, once impregnated, as the primary solid dielectric barrier.

Liquid mineral oil. The winding structure is submerged in refined electrical insulating oil. The oil serves two functions. It acts as a high-dielectric-strength medium that permeates the cellulose paper, and it acts as a convective heat transfer fluid that removes thermal energy from the winding coils.

Pressboard barriers and spacers. High-density pressboard elements establish physical isolation zones between phase windings, between windings and ground, and around structural components. These elements maintain defined clearance margins under all operating stresses, and they also set the clamping geometry that holds the winding stack in compression.

Engineering Trade-offs in Winding Insulation Design

Designing transformer winding insulation is a multi-variable optimization problem. Improving one performance dimension typically compromises another.

Dielectric margin against thermal dissipation. Increasing the thickness of paper wraps and pressboard structures allows a higher Basic Insulation Level (BIL) rating and greater lightning-impulse withstand. Thick cellulose also acts as a thermal blanket, trapping heat inside the conductor and raising the hot-spot temperature, which accelerates degradation.

Mechanical robustness against stacking factor. Thicker insulation boundaries better withstand the electromagnetic forces generated during a short circuit. They also increase the physical footprint of the windings, which reduces the copper stacking factor inside the core window, lowers power density, and raises manufacturing cost.

Viscosity against flash point. Lower-viscosity transformer oil improves convective cooling and penetration into the paper matrix. Lower-viscosity fluids typically carry a lower flash point, which raises fire risk in the event of arcing or severe local overheating.

Primary Causes of Insulation Loss

Winding insulation decomposition is driven by four interlinked stress vectors, commonly organized as thermal, electrical, ambient or chemical, and mechanical.

Thermal degradation. Continuous exposure to operational heat drives pyrolytic scission of the long polymer chains in cellulose, reducing degree of polymerization and, with it, the mechanical strength of the paper.

Electrical stress. Continuous exposure to nominal voltage, combined with high-frequency switching surges and lightning transients, erodes dielectric margins over time.

Ambient and chemical action. Moisture and oxygen act as catalysts. Moisture accelerates hydrolytic splitting of cellulose chains and generates organic acids, which further accelerate deterioration. Water is both a product of aging and an accelerant of it, and the rate of thermal deterioration of the paper is proportional to its water content. Ingress is substantially a weather process rather than an operational one. A free-breathing conservator draws ambient air in on every cooling cycle, and a depleted desiccant charge or a degraded seal turns each of those cycles into moisture accumulation. Sites with large diurnal swings and high humidity therefore accumulate water faster at identical electrical duty.

Mechanical stress. Cyclic thermal expansion, normal operational vibration, and mechanical shock from downstream through-faults cause micro-tears, loss of clamping pressure, and structural deformation in embrittled insulation.

The fourth vector matters disproportionately for what follows. Mechanical loosening changes the stiffness of the winding assembly, and stiffness is a mechanically observable quantity. Section 08 returns to this point.

Microscopic Mechanisms of Electrical Stress

Electrical stress degrades insulation through three microscopic mechanisms.

The void problem

Despite rigorous factory vacuum processing, microscopic gas bubbles can remain trapped within dense paper layers. Gas has a significantly lower permittivity than paper or oil, so in a series dielectric the local electric field concentrates inside these voids. Because gas also has a much lower breakdown threshold, it ionizes and produces localized micro-sparks known as partial discharge. Over time, partial discharge carbonizes the adjacent paper and creates a conductive path known as electrical treeing.

Micro-contaminants

Microscopic copper or steel burrs from manufacturing, together with suspended particulate matter, act as local electric field intensifiers. Under high AC voltage these particles polarize and align along the field lines, forming microscopic conductive bridges that lower the local breakdown voltage of the fluid insulation.

Sharp geometries

High-voltage fields concentrate at sharp boundaries, un-rounded conductor edges, and protruding structural hardware. These localized high-stress zones can exceed the dielectric strength of the surrounding fluid and initiate low-energy partial discharge. Sustained discharge cracks oil molecules over time, forming combustible gases and local carbon tracks.

Insulation Resistance Test Methodology

The Insulation Resistance test is the established offline diagnostic for evaluating winding insulation condition. The test applies a continuous DC voltage, typically 500 V to 5,000 V depending on winding voltage class, across the insulation structure under evaluation and measures the resulting current.

The total current measured during an IR test has three components.

Capacitive charging current. High initially, decaying rapidly toward zero as the winding structure charges.

Dielectric absorption current. Represents polarization of molecules within the dielectric. Decays slowly, over several minutes.

Conduction, or leakage, current. The steady-state current flowing through, or across the surface of, the insulation.

Technicians track the evolution of resistance over time and calculate the Polarization Index, defined in IEEE C57.152 clause 7.2.13.4 as the resistance at ten minutes divided by the resistance at one minute. A low ratio indicates that absorption current has been swamped by high steady-state leakage current, which is a warning sign of moisture saturation or contamination. The guideline bands for transformers are as follows [2].

Effectiveness and Limitations of Traditional IR Testing

The IR test is an effective coarse diagnostic and a recognized acceptance test. It also carries four structural limitations that constrain how it can be used.

Strong temperature dependence. Insulation resistance falls sharply as temperature rises, roughly halving for every 10 °C of increase. Readings must be corrected to a standard reference temperature before they can be compared, and the correction itself introduces error.

Blindness to localized defects. Because IR measures total leakage current across the entire insulation structure under test, a single severe localized defect, such as an advanced electrical tree or a mechanical fracture, can be masked by the surrounding healthy insulation.

Surface sensitivity. High ambient humidity or surface contamination on bushings creates superficial current paths, which can produce false indications of internal insulation deterioration unless guard-terminal configurations are used.

Indeterminacy on modern high-resistivity insulation. When the one-minute insulation resistance exceeds approximately 5,000 MΩ, total current falls into the sub-microampere range, where supply voltage variation, ambient humidity, and test lead conditions dominate the measurement. In that regime the calculated PI may not indicate winding condition at all. Modern transformers with high-resistivity oil commonly read above that threshold, so the PI calculation is often uninterpretable on precisely the units an owner most wants to trend.

There is a fourth consequence that follows from the first three. The acceptance floor in common use is one megohm per kilovolt of winding rating, which sets a threshold of 0.22 GΩ for a 220 kV winding and 0.069 GΩ for a 69 kV winding. In a healthy population, measured values clear that floor by one to three orders of magnitude. A pass-or-fail comparison against the floor therefore separates almost nothing, and the margin above it says little about how quickly the margin is being consumed.

Taken together, these limits describe a test that is valuable at the moments it is performed and silent in between. The interval, not the accuracy, is the exposure.

What VIE Measures

VIE does not perform an electrical test. It measures the mechanical state of the winding structure from outside the tank, and it interprets that measurement against a model of the specific unit.

The physical basis

A loaded transformer winding vibrates. Electromagnetic forces between current-carrying conductors act at twice the line frequency, and the winding stack, held in compression by its clamping structure and pressboard spacers, responds as a mechanical assembly with a characteristic stiffness. Core vibration travels predominantly through structure. Winding vibration travels predominantly through the oil. That difference in transmission path is what makes the winding contribution separable at the tank wall.

As insulation ages, the paper and pressboard lose thickness and elasticity, clamping pressure falls, and the effective stiffness of the assembly changes. Insulation aging is therefore not only a dielectric event. It is a mechanical one, and mechanical change is externally observable without penetrating the tank.

The three vibration observables

Amplitude. Responds to reduction in effective stiffness of the winding assembly.

Harmonic content. Responds to nonlinear contact stiffness, where the mechanical restoring force is no longer proportional to displacement.

Non-harmonic content. Responds to intermittent contact loss, where surfaces that should remain in continuous compression separate and re-seat within a cycle.

Each observable has a distinct physical mechanism behind it. The Radial and Axial Winding Health Metrics are derived from harmonic pattern analysis of the vibration signature, with the fundamental at twice line frequency. The metrics are not amplitude at a single frequency, and they are not a threshold on raw vibration level.

Four-signal fusion and the virtual model

Vibration is the anchor signal. It is fused with thermal measurement, magnetic field measurement, and local weather data. Sensors are external, passive, and battery powered, and they impose no voltage or current on the asset.

Local weather is a required input, not a convenience. Section 01 established that site environment moves the hot spot by an amount comparable to loading, and that ambient conditions drive the moisture and mechanical-cycling terms as well. The same environment acts on the measurement. Ambient temperature changes oil viscosity and therefore the transmission path from winding to tank wall. Solar gain and wind speed change the thermal gradient across the tank. A cold snap or a still, hot afternoon will move the vibration signature of a perfectly healthy transformer.

Without a weather channel, that movement is indistinguishable from a change in winding condition. With one, environmental effects can be gated out before the signature is compared against the model. Common-mode movement observed across every unit in a yard is treated as environmental. Movement confined to a single unit is treated as a candidate condition change. The fourth signal is what makes the first one trustworthy.

Interpretation is not a learned baseline. VIE constructs a physics-based virtual model from each transformer's approximate geometry and refines it with every sample. This distinction matters diagnostically. A learned baseline trained on a unit's own history will quietly absorb a slowly developing fault and re-normalize around it. A physics-based model will not, because the reference is what the geometry says the unit should do, not what the unit has recently been doing.

Why the metric relates to a dielectric test result

IR and PI report a dielectric quantity. The winding health metrics report a mechanical one. They are related because both respond to the same underlying degradation of the cellulose and pressboard system. Thermal and hydrolytic aging reduces the degree of polymerization of the paper, which reduces both its dielectric margin and its mechanical strength and thickness. The winding health metric degrades irreversibly under thermal stress, which is consistent with the irreversibility of the chemistry.

That is a mechanistic argument. Section 09 tests it against field measurements.

Field Correlation with Insulation Resistance

Section 08 argued from mechanism that a measurement of winding mechanical state should track a measurement of dielectric condition, because both respond to the same degradation of the cellulose system. This section tests that argument against field data.

The comparison set and the normalization method

Nine oil-filled transformers were compared. Two were units under continuous VIE monitoring in a customer pilot. Seven were reference units for which the asset owner supplied recent offline test results. The set spans four voltage classes from 33 kV to 231 kV and commissioning dates from 1995 to 2024, with two units for which the commissioning date was not recorded. For each unit, insulation resistance and Polarization Index were measured across three configurations: high voltage to ground, low voltage to ground, and high voltage to low voltage. The governing value used here is the lowest of the three.

Raw insulation resistance is not comparable across voltage classes, because the acceptance criterion scales with winding rating. At one megohm per kilovolt, a 231 kV winding must clear 0.231 GΩ and a 69 kV winding must clear 0.069 GΩ. Comparison across a mixed-class fleet therefore requires normalizing each measurement by its own acceptance floor. The resulting quantity, reported below as multiples of floor, is the margin the unit holds above the minimum it is required to meet.

Results

The radial winding health metric tracks insulation resistance margin, and the relationship is inverse, monotonic, and statistically significant at this sample size. Because insulation resistance margin spans nearly three orders of magnitude across the set, the linear relationship is against its logarithm rather than against the raw value, which is what the Pearson figures below are computed on.

What the results support, and what they do not

The radial metric is supported. Rank correlation against insulation resistance margin is -0.72 at p = 0.029, and the log-linear correlation is -0.86 at p = 0.003. Using each unit's maximum rather than its median strengthens both slightly. A unit with a lower radial metric holds more dielectric margin, across four voltage classes and three decades of commissioning dates.

The axial metric is not supported by this dataset. Rank correlation against insulation resistance margin is -0.51 at p = 0.160, which does not reach significance at n = 9. The axial metric responds to core and clamping structure geometry rather than to conductor insulation directly, so a weaker relationship to a dielectric measurement is expected. It should not be presented as an insulation resistance proxy.

No relationship to Polarization Index was found. Rank correlation is -0.31 at p = 0.415. This is a null result, and it is the result Section 07 predicts.

Age carries essentially no information. Across the seven units with recorded commissioning dates, rank correlation between the radial metric and age is -0.07 at p = 0.878. The oldest unit in the set, commissioned in 1995, sits in the healthy band on both measures. The newest, commissioned in 2024, sits mid-pack on the radial metric.

The Polarization Index null is a confirmation, not a gap

Section 07 states that when one-minute insulation resistance exceeds roughly 5,000 MΩ, total current falls into a regime where supply variation, humidity, and lead conditions dominate, and the calculated Polarization Index may not indicate winding condition at all. Seven of the nine units in this set measure above that threshold. The dataset contains the artifact directly: one unit returned a Polarization Index of 6.4 on one configuration and 1.5 on another, a spread that reflects measurement conditions rather than three different insulation systems inside one tank.

A method that correlated cleanly with Polarization Index across this population would be tracking measurement noise. The absence of that correlation is what the physics predicts, and finding it is a check on the method rather than a limitation of it.

The case that both methods flag

Unit 4 is the anchor. It holds the lowest dielectric margin in the set by a wide margin, clearing its acceptance floor by a factor of six where every other unit clears by 34 to 1,787. Its Polarization Index of 1.09 falls in the poor band of Table 1, adjacent to dangerous. Its radial winding health metric is 3.65, more than double the next highest unit in the set.

Two independent measurements, one dielectric and offline, one mechanical and continuous, identify the same asset as the outlier. Unit 6 is the second-ranked concern on both. Neither measurement was calibrated against the other.

Limits of this dataset

Sample size. Nine units. The radial finding is significant at this n, but the axial and Polarization Index nulls are underpowered. Absence of a detectable relationship at n = 9 is not evidence that no relationship exists.

Test conditions not controlled by VIE. Insulation resistance results were produced by the asset owner's own maintenance program, not under a protocol designed for this comparison. Temperature correction was applied by the testing party and has not been independently verified.

Commissioning dates incomplete. Two of nine units have no recorded commissioning date, so the age analysis rests on seven.

The claim this dataset supports is bounded accordingly. The radial winding health metric tracks insulation resistance margin across a mixed-class population, independent of asset age.

Continuous Monitoring Compared with Periodic Testing

The two methods answer different questions and are not interchangeable. Periodic dielectric testing establishes whether the insulation meets a withstand criterion on the day of the test. Continuous monitoring establishes whether the mechanical condition of the winding is changing between those days. The operational profile differs across every dimension that matters to an asset owner.

The operational friction of periodic testing

Despite its diagnostic value, IR testing is performed rarely, often years apart, for three reasons.

Mandatory de-energization. The asset must be powered down and isolated from the grid, which causes operational disruption, costly switching procedures, and lost availability.

Test-induced stress. Applying DC voltage to aged insulation carries a non-zero probability of accelerating degradation or precipitating a failure during the test itself.

Labor-intensive interpretation. Raw IR data requires expert manual normalization, temperature correction, and review, which introduces delay and analyst-to-analyst variation.

None of these are objections to the test. They are the reason the interval between tests is measured in years, and the reason condition can change substantially inside that interval without anyone observing it.

Scope and Limitations of the Method

Stating what the method does not do:

It is not a dielectric withstand test. VIE does not measure insulation resistance, dielectric strength, or breakdown voltage, and it does not produce a number that can be substituted for one.

It does not replace acceptance or commissioning testing. IR, PI, power factor, and turns ratio testing remain the recognized instruments for acceptance, commissioning, warranty, and post-fault verification under IEEE C57.152.

It does not locate a defect geometrically. The winding health metrics are aggregate per monitored unit. They resolve change over time. They do not identify which turn or which disc has deteriorated.

It does not detect every failure mode. Failure modes with no mechanical signature, and modes originating outside the winding structure, are outside what the vibration observables can report.

The axial metric is not an insulation proxy. Section 09 found no significant relationship between the axial metric and insulation resistance at n = 9. The axial metric reports on core and clamping geometry. It should not be offered as a substitute for a dielectric measurement.

Coverage means temporal coverage. Continuous monitoring closes the gap between tests. It does not claim greater diagnostic depth at any single instant than a dedicated offline test performed correctly.

Periodic testing verifies a condition against a standard. Continuous monitoring reports whether that condition is holding.

Conclusion

Winding insulation is the effective limiting factor on transformer life, and its degradation is driven by well understood thermal, electrical, chemical, and mechanical mechanisms. The Insulation Resistance test has been the industry standard for evaluating that degradation for decades, and it remains the recognized acceptance instrument. Its reliance on de-energization, its potential to stress the asset under test, its dependence on manual interpretation, and its documented indeterminacy on high-resistivity modern insulation mean it is applied infrequently and cannot report condition between applications.

The same body of field evidence that establishes winding failures as the leading contributor to major transformer failures also shows that age alone is a weak predictor of which unit will fail. What separates two identical units is accumulated thermal, chemical, and mechanical exposure, and that exposure is set jointly by how the unit has been worked and by where it stands. Duty and site are both multipliers, and neither is recorded on the nameplate. That makes condition a per-unit, in-place measurement problem rather than a fleet actuarial one.

Continuous, non-invasive, physics-based monitoring addresses the interval rather than the instrument. It observes the mechanical consequence of insulation aging on an energized asset, under real load and real ambient conditions, at a sampling rate that matches the rate at which condition can actually change.

The field data in Section 09 supports the mechanistic argument for doing so. Across nine transformers spanning four voltage classes, the radial winding health metric tracks insulation resistance margin at a rank correlation of -0.72, and does so independently of asset age. Both methods independently identify the same unit as the population outlier, without either having been calibrated against the other. The dataset is small and cross-sectional, and it does not yet demonstrate the stronger longitudinal claim. What it does establish is that the mechanical observable and the dielectric one are reporting on the same underlying condition. That is what makes continuous monitoring a substitute for the silence between tests rather than a substitute for the tests. It does not replace the dielectric test. It removes the years in which nobody is watching.

References

1. CIGRE Working Group A2.62. Analysis of AC Transformer Reliability. CIGRE Technical Brochure 939, September 2024. Population of more than 425,000 transformer-years, 1,204 major failures and 1,916 retirements, contributed by 66 utilities in 27 countries.
2. IEEE Std C57.152-2013. IEEE Guide for Diagnostic Field Testing of Fluid-Filled Power Transformers, Regulators, and Reactors. Clause 7.2.13.4 defines the Polarization Index and its interpretation bands.
3. IEEE Std C57.91-2011. IEEE Guide for Loading Mineral-Oil-Immersed Transformers and Step-Voltage Regulators. Defines the aging acceleration factor and the thermal dependence of insulation life.
4. IEC 60076-7:2018. Power transformers, Part 7: Loading guide for mineral-oil-immersed power transformers. Defines the hot-spot temperature as the sum of ambient temperature, top-oil rise, and hot-spot rise, and tabulates the relative aging rate of cellulose as a function of temperature, moisture content, and oxygen content.
5. VIE Technologies Inc. Continuous monitoring pilot report, December 2025. Source of the nine-unit comparison set in Section 09, comprising two units under continuous monitoring and seven reference units with asset-owner offline test results. Unit identifiers anonymized.