How to interpret a DGA report: Key Gas, Rogers ratios and Duval triangles

A dissolved gas analysis (DGA) report lists a dozen gases in parts per million and usually ends with a one-line comment from the lab. If one number jumped, you are left deciding whether to resample, keep running, or plan an outage. The interpretation methods exist to turn those numbers into a fault type. This article explains how the three most used ones work (the Key Gas method, Rogers ratios and the Duval triangles), where each one misleads, and how to read them together.

Where the gases come from

Mineral oil and paper break down under heat and electrical stress, and the pieces dissolve in the oil as gas. Which gases form depends mostly on how much energy the fault puts into a small volume:

• Hydrogen (H2) forms at the lowest energies. Partial discharge in oil produces mostly hydrogen.
• Methane (CH4) and ethane (C2H6) come from oil heated to a few hundred degrees.
• Ethylene (C2H4) takes over as the oil gets hotter, and dominates serious thermal faults.
• Acetylene (C2H2) needs the highest temperatures, which in practice means arcing or an extremely hot spot.
• Carbon monoxide (CO) and carbon dioxide (CO2) come mainly from paper (cellulose) insulation. Normal aging makes both, so the ratio between them and how fast they rise matter more than the level.

Oxygen and nitrogen are not fault gases, but they tell you how the oil is preserved and whether the sample was exposed to air. Every method below works from these same few gases. They differ in which ones they use and how they compare them.

Before any method: is the sample real?

More bad DGA calls start with a bad sample than with a bad method. Check these before reading any diagnosis:

• A jump that the next sample does not confirm. A real fault rarely vanishes on its own. If a high result is followed by a normal one, the high sample was probably contaminated, and a licensed engineer should still review it before it is written off.
• Oxygen and nitrogen that do not fit. In a sealed or nitrogen-blanketed unit, a sample with much more oxygen than its neighbours was probably exposed to air in the syringe or the bottle.
• Two transformers with the same numbers on the same day. Units with different histories almost never produce matching gases. That pattern points to a mix-up or contaminated sampling equipment.
• Oil processing. Degassing, filtering or an online oil conditioner resets or suppresses the gas levels. Trends have to be read from the processing date forward.
• Levels near the detection limit. Ratios of very small numbers mean nothing. As a working rule, I do not lean on a ratio method until the gases in it are roughly ten times the lab's detection limit.

The rate of change also deserves more weight than any single level. A gas that has sat at the same moderate value for ten years is less worrying than one that doubled since the last sample.

The Key Gas method

The Key Gas method looks at which gas dominates and names the fault from it. Mostly hydrogen suggests partial discharge. Mostly ethylene suggests overheated oil. Acetylene with hydrogen suggests arcing. Large amounts of carbon monoxide suggest overheated paper.

IEEE C57.104 pairs this with limits for each gas and for total dissolved combustible gas (TDCG). The 2008 edition sorts results into four conditions with sampling intervals and actions for each. The 2019 revision replaced those with status levels that also account for the transformer's age, how the oil is preserved, and the change between samples. Read your own copy for the values; this page does not reproduce them.

The method is fast and easy to explain to a plant manager. Its weakness is that it only looks at the biggest gas. Two faults at once, or a fault developing on top of normal aging, can produce a mix that points nowhere in particular. TDCG has its own trap: one gas can climb into a high condition while the total stays low, which is why the individual gas limits matter.

Rogers ratios and the IEC three-ratio code

Ratio methods compare gases that form at neighbouring temperatures, so the answer depends on the shape of the gas mix and not its size. Rogers built his method from a large set of transformers with known faults, refining Doernenburg's earlier ratio method. The version in IEEE C57.104 uses three ratios:

• Methane to hydrogen (CH4/H2) separates discharge from heating.
• Acetylene to ethylene (C2H2/C2H4) separates arcing from thermal faults.
• Ethylene to ethane (C2H4/C2H6) rises with fault temperature.

Each ratio falls into a range, the ranges combine into a short code, and the code maps to a case such as partial discharge, low-energy arcing, or a thermal fault in a given temperature band. IEC 60599 uses the same three ratios with its own ranges and codes, which is why you will see a three-digit IEC code on some lab reports.

Rogers ratios are good at grading thermal faults by temperature. Their weakness is the gaps: some ratio combinations match no case at all, and the method then has nothing to say. They also need all five gases well above detection, or a single low reading in a denominator can flip the code.

Duval triangles

Michel Duval's method takes three gases, converts each to a percentage of their sum, and plots the result as one point on a triangle. The triangle is divided into zones, each tied to a fault type. Because every point lands in some zone, the triangle always gives an answer, which is both its strength and the reason to use it only when the gases are actually abnormal.

Triangle 1 is the general one for mineral oil. It uses methane, ethylene and acetylene, and its zones are:

• PD, partial discharge
• D1, low-energy discharge (sparking)
• D2, high-energy discharge (arcing)
• DT, a mix of thermal and electrical faults
• T1, thermal fault below 300 °C
• T2, thermal fault from 300 to 700 °C
• T3, thermal fault above 700 °C

Other triangles cover cases Triangle 1 cannot. Triangle 2 is for load tap changer compartments, where some arcing is normal. Triangle 3 is for natural ester, silicone and other non-mineral fluids, which gas differently. Triangles 4 and 5 refine a mineral-oil diagnosis once Triangle 1 has found a fault. Triangle 4 uses hydrogen, methane and ethane to split low-temperature results into partial discharge, stray gassing of the oil, overheating, and possible paper carbonization. Triangle 5 uses methane, ethylene and ethane to do the same for higher-temperature faults.

Duval later added two pentagons that use five gases at once (hydrogen, methane, ethane, ethylene and acetylene). They cover the same fault families and also catch stray gassing, which the original triangle could not separate from a real fault.

When the methods disagree

They often do, especially when gas levels are modest or two faults overlap. A sensible order is:

1. Confirm the sample, using the checks above.
2. Decide whether anything is abnormal, from the IEEE levels and above all from the rate of change.
3. Only then run the fault-type methods, and look for two or more that agree on the family of fault (electrical or thermal) and roughly on its energy.
4. Where they disagree, trust the trend over the single sample, and resample sooner rather than later.

A real case: a step-up transformer

I reviewed the DGA history of a generator step-up transformer at a combined-cycle plant. The sequence is a good example of the methods working, and of where they stop.

The first warning came from routine sampling, which showed a low-temperature thermal fault and gas levels high enough that the owner took the unit out of service. An internal inspection found loose bolts on internal bus connections. They were tightened, the oil was processed, an online oil conditioner was installed to keep stripping gas, and the unit went back into service. Gas stayed low for about two weeks. Within two months it was climbing again.

By the time the unit was removed for good, TDCG still sat in the lowest condition, because the oil conditioner was pulling gas out as fast as the fault made it. The individual gases told a different story: ethylene was in Condition 3 and heading for Condition 4, with methane and ethane elevated. Every method pointed the same way:

• Rogers ratios gave a thermal fault between 300 and 700 °C.
• The Key Gas method pointed to overheated oil.
• Duval Triangle 1 plotted on the boundary between T2 and T3.
• Triangle 5 plotted between T3 and C, possible carbonization of paper.

When the tank was drained and opened, the cable from the tap changer to one of the high-voltage winding taps was found badly overheated. The paper on it was carbonized black, the copper strands crumbled when touched, and about three quarters of them had already parted in service. The transformer was close to failing. The lead was crimped into a splice at the winding end and had two cables in one lug at the tap changer end, both likely sources of a high-resistance joint.

Three lessons from this unit apply to most DGA histories:

• With online gas removal in service, TDCG can sit in the lowest condition while an individual gas climbs toward the worst one. Read each gas, and read the trend.
• Four methods agreeing on a hot thermal fault, with Triangle 5 adding paper involvement, was strong evidence of a real fault, and the inspection confirmed it.
• DGA named the type of fault correctly and could not say where it was. When gas returns after a repair, the repair may not have found the source.

What DGA cannot tell you

DGA finds faults that make gas. It does not locate them, and normal gas levels do not prove a transformer is free of problems: some failure modes give little warning in the oil. Treat DGA as one input next to the unit's loading, its electrical test history, its age, and how it has been maintained. The final call on an outage belongs to the owner and their engineer.