Insulation Diagnostics and Condition Monitoring

A withstand test asks a blunt question: does the insulation survive a defined overvoltage, yes or no? It is a gate — pass and the apparatus ships, fail and it does not. That is exactly what is wanted when a new machine leaves the factory. But the great population of high-voltage plant in service — transformers, bushings, cables, switchgear and rotating machines, much of it decades old — cannot be answered with a single gate. What an asset owner needs is not a verdict at one instant but a trajectory: how the insulation is ageing, what is driving the decline, and how much useful life remains. That is the realm of diagnostics and condition monitoring.

The shift is one of philosophy. Reactive maintenance runs equipment to failure, which is cheap until the failure is catastrophic. Time-based preventive maintenance services everything on a fixed calendar, which is safer but wasteful, since healthy units are disturbed needlessly while a sick one may fail between visits. Condition-based maintenance threads between the two: each asset is measured, its condition trended, and intervention is timed to the evidence. The diagnostic tests of this part — partial discharge, tan δ, and the methods below — are the instruments that make condition-based maintenance possible, because each turns an invisible internal state into a number that can be watched.

The simplest and oldest diagnostic is the insulation resistance (IR) test: a DC voltage (commonly \(500~\mathrm{V}\) to \(5~\mathrm{kV}\)) is applied with a megohmmeter and the resulting current — hence the resistance — is read. What makes it diagnostic rather than a simple ohm reading is the way the current changes with time. When DC is suddenly applied to insulation, three currents flow together:

The capacitive current charges the geometric capacitance and dies away in seconds. The absorption (polarization) current decays much more slowly, over minutes, as the dielectric's slow polarization mechanisms settle. The conduction current is the small steady leakage that remains. In dry, clean insulation the absorption current keeps falling for many minutes, so the measured resistance keeps rising; in wet or contaminated insulation a large, constant leakage swamps the decaying terms, and the resistance is high from the start but never climbs. Two simple ratios capture exactly this behaviour:

As a rough guide for machine windings, a polarization index above about \(2\) indicates dry insulation, while a value near \(1\) calls for drying or cleaning before the machine is returned to service. The same logic — watch how the charging settles, not just where it starts — underlies the more sophisticated dielectric-response methods that follow.

The polarization index reads the dielectric's slow response at two coarse instants. Dielectric-response methods generalise this into a full curve, and from that curve they extract one prize above all others for oil-paper insulation: the moisture content of the paper, which governs both its dielectric strength and the rate at which it ages. Three methods are in use, two in the time domain and one in the frequency domain.

In the polarization–depolarization current (PDC) method, a DC voltage is applied for a long time while the slow polarization current is recorded; the supply is then removed and the specimen short-circuited, and the depolarization current is recorded as the stored polarization relaxes. The shapes of these two slow currents encode the conductivity of the oil and the moisture and ageing of the paper. The older recovery voltage method (RVM) works similarly but measures the voltage that re-appears across an initially charged, then briefly shorted, specimen, mapping out a "polarization spectrum" whose central time constant was once used as a moisture indicator.

The modern favourite is frequency-domain spectroscopy (FDS), also called dielectric frequency response. Here the loss tangent and capacitance are measured across a wide band — typically from a few millihertz up to a kilohertz — turning the single tan δ point of Chapter 20 into a full spectrum. Moisture and ionic conduction dominate at the lowest frequencies, lifting the loss steeply there, while the bulk geometry sets the high-frequency end. By fitting the measured curve against a library of oil-paper responses, the moisture in the solid insulation can be estimated without taking the apparatus apart. This is the diagnostic descendant of the tan δ measurement: the same loss angle, now read as a function of frequency to separate the causes that a single number blends together.

For oil-filled apparatus — above all the power transformer — the single most powerful diagnostic is dissolved gas analysis (DGA). Its principle is beautifully direct: whenever oil or paper is stressed beyond a threshold, by local overheating or by electrical discharge, the molecules break apart and release characteristic gases, which dissolve in the oil. A small oil sample, degassed and run through a gas chromatograph, then reveals — in parts per million — a chemical fingerprint of whatever fault is at work, often long before any electrical symptom appears.

The diagnostic gases fall into a few families. Hydrogen \((\mathrm{H_2})\) accompanies almost any disturbance and is especially associated with partial discharge. The light hydrocarbons — methane \((\mathrm{CH_4})\), ethane \((\mathrm{C_2H_6})\), ethylene \((\mathrm{C_2H_4})\) — appear in a sequence that climbs with temperature, ethylene marking the hotter thermal faults. Acetylene \((\mathrm{C_2H_2})\) is the alarm gas: it forms only in the intense heat of arcing, so its presence in any quantity signals a high-energy electrical fault. Finally, the carbon oxides — carbon monoxide \((\mathrm{CO})\) and carbon dioxide \((\mathrm{CO_2})\) — come from the cellulose of the paper, so they report on the involvement and overheating of the solid insulation.

Just as important as which gases are present is how fast they are growing. A modest but steadily rising concentration is more worrying than a high but stable one, since the rate of generation (in ppm per day) measures how active the fault is right now. Standards such as IEEE C57.104 set both absolute concentration levels and rate-of-change limits, escalating the response from routine sampling, through increased surveillance, to removal from service as the total combustible gas and its growth rate climb.

A list of gas concentrations is raw data; turning it into a fault diagnosis takes an interpretation scheme. The key-gas method simply reads off the dominant gas. More robust are the ratio methods — the Rogers ratios and the IEC 60599 scheme — which take the quotients of selected gas pairs, because ratios are less sensitive than absolute amounts to the size of the fault and the volume of oil. Three ratios do most of the work:

The pattern of these three places the fault into a category: a partial discharge, a low-energy or high-energy electrical discharge, or a thermal fault of low, medium or high temperature. A large \(R_1\) (acetylene present) points to arcing; a large \(R_5\) (ethylene over ethane) points to a hot thermal fault; \(R_2\) helps separate discharge from heating. The ratio schemes are powerful but have a known weakness: some gas combinations fall outside every defined code, leaving no diagnosis.

The Duval triangle was devised to close that gap. It uses just the three hydrocarbon gases — \(\mathrm{CH_4}\), \(\mathrm{C_2H_4}\) and \(\mathrm{C_2H_2}\) — expressed as percentages of their sum, so every possible sample maps to exactly one point inside an equilateral triangle. The triangle is partitioned into zones, and the zone containing the point names the fault. Because every point lands somewhere, the triangle always returns an answer.

Ratios and triangle are best used together and alongside the gas trend: a consistent diagnosis from several schemes, confirmed by a rising generation rate, is far stronger evidence than any one method alone. This is the recurring theme of the chapter — no single indicator is trusted on its own.

Dissolved gases report on active faults; the quality of the oil itself reports on the slow ageing of the insulation system. Routine oil tests measure the breakdown voltage (which falls with moisture and particles), the water content by Karl Fischer titration, the acidity or neutralization number (which rises as the oil oxidises), the interfacial tension (which falls with the same oxidation), and the oil's own dissipation factor. Together these say whether the oil is still a sound dielectric and heat-transfer medium, or whether it should be reconditioned or replaced.

But the oil can be changed; the paper cannot. The cellulose of the solid insulation is the true life-limiting element of a transformer, and as it ages its long molecular chains break, measured by the degree of polymerization (DP) — about \(1000\)–\(1200\) when new, falling toward \(150\)–\(200\) at end of life, where the paper has lost most of its mechanical strength. DP can be measured directly only on a paper sample, which usually means cutting into the machine. The elegant alternative is furan analysis: as cellulose degrades it releases furanic compounds, chiefly 2-furaldehyde (2-FAL), into the oil, where it can be measured from the same kind of sample as DGA. An empirical correlation then turns the 2-FAL concentration into an estimate of DP — and hence of remaining paper life — without opening the tank.

The carbon-oxide ratio from the DGA helps too: a \(\mathrm{CO_2/CO}\) ratio that drifts low can corroborate accelerated paper degradation. As always, furan, DP and the carbon oxides are read together and over time — the paper's ageing is a slow trend, and it is the trajectory toward the end-of-life DP that an asset owner is really tracking.

Everything so far is, in its classic form, an offline test — the apparatus is sampled or de-energised, measured, and the result trended visit by visit. The modern direction is to bring the measurement online, with permanent sensors that watch the asset continuously and stream their readings to a monitoring system. Several of the diagnostics map naturally onto sensors:

The bushing tap gives a continuous reading of capacitance and tan δ, catching a developing bushing fault — historically a leading cause of transformer failure. High-frequency current transformers and UHF antennas pick up partial discharge while the unit stays energised. Online gas sensors track hydrogen, and increasingly the full gas suite, between laboratory samples. Fibre-optic probes read winding and hot-spot temperature directly, where once it could only be inferred. Fed into a monitoring platform, these streams move the asset from periodic snapshots to a living record — the difference between checking a patient yearly and wearing a continuous monitor. The cost is justified on the most critical, most expensive or most loaded units, where an unexpected failure is least affordable.

The chapter's recurring lesson is that no single test tells the whole story. Partial discharge finds local defects but misses diffuse ageing; tan δ weighs the bulk but cannot locate a fault; DGA reads active faults but not mechanical strength; furan reads paper life but not an incipient discharge. The art of diagnostics is data fusion — combining many indicators, each with its blind spot, into a verdict more reliable than any one of them.

In practice this is formalised as an asset health index: each diagnostic is scored against limits and trends, the scores are weighted by how strongly each predicts failure, and the weighted aggregate places the asset on a scale from healthy to end-of-life. The index is not a precise physical quantity but an engineering judgement made repeatable, and its real output is a decision — continue in service, increase surveillance, refurbish, or replace — and a ranking of a whole fleet so that limited maintenance budgets flow to the units that most need them. This is where high-voltage diagnostics meets asset management: the physics of the earlier chapters becomes, in the end, a defensible plan for keeping the lights on.

For each item, first identify what it tests — the diagnostic philosophy, an index, a dielectric-response idea, the DGA gases, a ratio or triangle interpretation, or paper ageing — then apply it. Difficulty rises down the list.