Water in Transformer Oil: Why a Low ppm Result Does Not Mean a Dry Transformer
Last verified: 2026-08-01
A transformer can return a reassuring water-content number and still be wet. The routine test extracts a small oil sample and titrates it for moisture — but in a working transformer, most of the water is not in the oil. It is bound inside the cellulose insulation wrapped around the windings, and the oil figure only estimates that hidden reservoir indirectly, through a relationship that shifts with temperature. A low ppm result on the day of sampling does not rule out a unit whose paper insulation is carrying a meaningful moisture load.
A water-content result describes the oil, not the transformer.
Where does the water actually sit?
Two things the ppm figure hides are worth naming directly. Most of the system's water sits in the solid insulation, not in the liquid the lab actually tests. And the number that is reported moves with temperature, because oil's capacity to hold water in solution is itself temperature-dependent.
Cellulose paper and pressboard are strongly hygroscopic — they attract and hold water far more readily than mineral oil does. Mineral oil, by contrast, dissolves only a small quantity of water before it reaches saturation, the point at which additional moisture stops staying in solution. An oil sample, then, is not a portrait of the transformer's moisture. It is an indirect proxy for a much larger reservoir held elsewhere.
Why do the published oil-to-paper ratios disagree?
Search for how water divides between paper and oil and several ratios turn up, presented as if they were the same fact. They are not. Published sources describe the relationship in at least three different ways: as an affinity ratio, describing how strongly each material attracts water relative to the other; as a concentration ratio, describing water per unit mass of each medium; and as a total-mass ratio, describing how much water each medium holds in a specific tank — which depends on how much paper and how much oil that particular tank contains.
These are three distinct physical quantities, and the specific proportions attached to them vary between sources without a traceable common primary. That is not evasion; it is a structural fact. A mass ratio depends on the paper-to-oil mass of the individual unit being discussed, so no single universal number can describe every transformer's split — a small distribution unit and a large power transformer do not share a paper-to-oil mass ratio, and neither figure generalises to the other. What is well documented, and consistent across sources, is the direction: at equilibrium, the solid insulation holds the large majority of the system's water. The magnitude simply is not one number, and treating it as one is the more common error.
How can the same ppm number mean two different things?
Two transformers can report identical absolute water content in ppm and sit in very different states, because ppm alone says nothing about how close the oil is to its saturation limit at that moment. Relative saturation compares the measured water content against the oil's saturation ceiling at the sampling temperature — and that ceiling itself moves with temperature, fluid type, and the oil's own age and oxidation state.
This is why relative saturation, not absolute ppm, is the physically comparable quantity across readings taken at different times or temperatures. CIGRE Technical Brochure 741, produced by Working Group D1.52 in 2018 and titled Moisture measurement and assessment in transformer insulation — Evaluation of chemical methods and moisture capacitive sensors, examines how absolute Karl Fischer readings convert to relative-saturation figures, and it addresses directly the uncertainties involved in that conversion.
If a report lists only an absolute ppm figure with no sampling temperature attached, then that figure cannot be placed on a saturation scale at all. It is a measurement missing the context needed to interpret it.
Can a lab back-calculate paper moisture from an oil reading?
Yes, but only under an assumption a loaded transformer rarely satisfies. Estimating how much moisture sits in the paper from an oil reading relies on published moisture-equilibrium curves — most notably T.V. Oommen's 1984 IEEE paper, Moisture Equilibrium Charts for Transformer Insulation Drying Practice, and CIGRE Technical Brochure 349 (Working Group A2.30, 2008), Moisture Equilibrium and Moisture Migration within Transformer Insulation Systems.
Here is the nuance almost nothing on the open web states plainly: those curves assume equilibrium between the paper and the oil. Equilibrium means water has stopped migrating between the two media, and that condition does not hold while temperature is changing — which is the normal operating state of a loaded transformer. Moisture migrates between paper and oil as the unit heats and cools, and that migration takes time to settle.
Equilibrium is a laboratory condition. A working transformer rarely holds still long enough to reach it. A paper-moisture estimate derived from an oil sample is therefore at its least reliable exactly when the unit is doing its job.
What else shifts the saturation ceiling?
Two more variables move where the saturation ceiling sits, independent of how much water is actually present.
As mineral oil oxidises in service it accumulates polar degradation products, and those products raise the oil's affinity for water. The practical effect: the same ppm figure represents a lower relative saturation in aged, oxidised oil than it would in new oil. The oil has effectively become more absorbent even though the absolute reading looks unchanged.
Fluid type matters more. Ester fluids dissolve substantially more water than mineral oil, because the ester molecule is polar where a hydrocarbon molecule is not. The standards themselves reflect this: natural ester fluids are specified under IEC 62770:2024 and synthetic esters under IEC 61099, held separately from mineral oil under IEC 60296:2020, and the ester specifications permit materially higher water content than the mineral-oil specification does. Mineral-oil ppm intuition does not transfer to an ester-filled unit — a figure that would flag a mineral-oil transformer can be unremarkable in an ester-filled one.
How is the water actually measured?
Karl Fischer titration is the standard laboratory method behind most absolute water-content figures reported in this field. IEC 60814:1997 (Edition 2.0), issued by IEC Technical Committee 10, specifies automatic coulometric Karl Fischer titration for water content in insulating liquids and — distinctly — in oil-impregnated cellulosic insulation; its scope reaches the solid as well as the liquid.
ASTM D1533-20, approved 1 November 2020, is titled Standard Test Method for Water in Insulating Liquids by Coulometric Karl Fischer Titration. Its scope, as the current title states, is coulometric and addresses the liquid; it should not be described as covering a volumetric method or an in-situ solid-insulation measurement, because neither is supported by its present title.
One distinction is worth carrying forward: one standard's scope extends into the solid insulation, the other's does not.
Does the sample itself introduce error?
Yes — sampling method, and even the citation used, can be a source of error. Two live examples make the point better than any general warning.
ISO 3170:2025 is titled Hydrocarbon Liquids — Manual sampling; earlier editions carried the title Petroleum liquids — Manual sampling. A specification still citing the old title is pointing at a superseded document, worth catching before it is repeated into a purchase spec.
The sampling standard for insulating liquids carries the same trap twice over. The current document is IEC 60475:2022 (Edition 3.0), Method of sampling insulating liquids, which cancels and replaces the second edition of 2011 — and the older title Method of sampling liquid dielectrics belongs to the first edition, retired long before that. Its scope applies to liquids below 1,500 mm²/s at the sampling temperature and covers mineral oils alongside non-mineral liquids including natural esters, synthetic esters and silicones, so an ester-filled unit is sampled under the same document. ASTM D923-15(2023), titled Standard Practices for Sampling Electrical Insulating Liquids, carries its (2023) suffix because it was reapproved without technical change that year — cite it with the suffix, not as a 2015-only document.
One boundary deserves stating explicitly, because it gets blurred in practice. Dissolved gas analysis is a different test: gas extraction and analysis are governed by IEC 60567:2023, while the oil-sampling step itself is specified in IEC 60475:2022. DGA characterises fault gases dissolved in the oil. It is not a moisture check, and a report that treats one as covering the other has merged two separate diagnostic disciplines.
What counts as an acceptable result?
There is no single ppm threshold that applies across transformers, and treating one as universal is itself a source of error.
IEC 60422:2024 (Edition 5.0), titled Mineral insulating oils in electrical equipment — Supervision and maintenance guidance, is the document governing in-service oil supervision. It is scoped to oil originally supplied to IEC 60296, and — worth stating plainly — it frames itself as a basis for an operator to write its own local code of practice, not as one universal pass/fail table. IEC 60296:2020 is a different kind of document: a delivery-acceptance specification for unused oil, applied before the oil ever goes into service. The two are routinely confused, and they answer different questions.
What actually decides what is acceptable for a given transformer is a set of variables, not a fixed number: the voltage class and equipment type; whether the insulation system is sealed or free-breathing; the oil's own condition and oxidation state; the unit's loading history and operating temperature; and the operator's own written specification or local code. Naming those variables, and the document that frames them, is the honest answer to what the limit is — because the limit itself is not one number.
Why does any of this matter physically?
Moisture is not a bookkeeping concern; it changes what the insulation can withstand. Dielectric strength of the oil-paper system falls as relative saturation rises, which is why breakdown-voltage testing under IEC 60156 is read alongside a moisture result rather than in isolation from it.
The sharper risk shows up under thermal overload. Moisture held in the paper can be driven out as vapour bubbles at the winding surface, and those bubbles collapse dielectric strength exactly where the insulation is under the most electrical stress — a mechanism documented by T.V. Oommen and S.R. Lindgren in their 2001 IEEE PES conference paper, Bubble evolution from transformer overload, and addressed within the scope of CIGRE TB 349.
Moisture also accelerates acid-hydrolysis ageing of cellulose, reducing its degree of polymerisation over time. The paper becomes mechanically weaker as this proceeds, independent of any single electrical event. None of these consequences are visible in a ppm figure read on its own.
Which details turn a number into a diagnosis?
A water-content figure is only interpretable alongside its own context.
What must accompany a water-content result on an insulating-oil report
- The test method and its edition — for example IEC 60814:1997 or ASTM D1533-20
- The sampling standard used — for example ISO 3170:2025, IEC 60475:2022, or ASTM D923-15(2023)
- The oil temperature at the moment of sampling
- The fluid type — mineral oil, natural ester, or synthetic ester
- Whether the figure reported is absolute water content in ppm, or relative saturation as a percentage of the oil's saturation limit at the stated temperature
A result missing any one of these five items is a number without a frame. It can be recorded, but it cannot yet be judged — and if the sampling temperature is absent, the figure cannot be converted to relative saturation at all, which makes it uncomparable with every other reading on the same unit.
Require those five items in writing rather than the ppm number alone, whether the report arrives at delivery, at acceptance testing, or as part of ongoing supervision. A buyer sourcing insulating oil through the RFQ process on Altonex Global — or sourcing an independent third-party testing service the same way — can write that requirement into the request itself, so suppliers return an interpretable result instead of an isolated figure.