Dissolved gas analysis (DGA) is still the most practical way to look inside an oil-filled power transformer, and the tools around it keep improving. In late August 2026 a machine-learning study published in ELECTRICA reported up to 98% classification accuracy on real service-transformer DGA datasets, and a hybrid model in IEEE Transactions on Power Delivery reached 94.68% on the IEC TC-10 database. Those are published research results, yet a fault is rarely caught by a model alone: someone must read the oil report, judge the trend, and decide whether to act. This dga interpretation guide lays out a hydrogen-first workflow built for an engineer on the substation floor rather than a research dataset.
Hydrogen is where interpretation begins
Hydrogen (H₂) is normally the first gas released when an internal abnormality develops. Its small molecule diffuses out of hot oil zones early, and it is produced across the full fault temperature range, from low-energy partial discharge up to thermal faults above 700 °C. That makes dissolved hydrogen the earliest practical signal for overheating, arcing, and insulation stress alike.
Most offline DGA programs take oil samples every three to six months. A slowly forming defect can cross a meaningful hydrogen threshold and sit there between samples; the first interpretation problem is timing, not complexity. Treat hydrogen as the tripwire, then use the rest of the gas picture to identify the fault.
A five-step workflow for interpreting DGA results
A usable dga interpretation guide turns a table of gas concentrations into a short list of actions. The sequence below condenses common utility practice and the decision logic of IEEE C57.104-2019, and it scales from a single report to a fleet review.
| Step | What you check | What it tells you |
|---|---|---|
| 1 | Hydrogen level and its first increase | Earliest sign of thermal or electrical stress |
| 2 | Key gas combination | Fault family: partial discharge, thermal, or discharge |
| 3 | Ratio or Duval zone | Refined fault type (PD, D1, D2, T1-T3) |
| 4 | Rate of change between samples | Active fault vs. stable condition |
| 5 | CO/CO₂ and moisture cross-check | Whether paper insulation is involved |
Key fault gases, ratios and Duval methods
Seven gases do most of the interpretation work. Hydrogen leads the sequence; the hydrocarbons add temperature and energy context; the carbon oxides point at the paper insulation.
| Gas | Typical source | Main indication |
|---|---|---|
| Hydrogen (H₂) | Oil cracking, partial discharge | Earliest warning; PD or thermal stress |
| Methane (CH₄) | Thermal fault below 300 °C | Low-temperature thermal activity |
| Ethane (C₂H₆) | Thermal fault 300-700 °C | Mid-temperature oil decomposition |
| Ethylene (C₂H₄) | Thermal fault above 700 °C | Hot-spot formation |
| Acetylene (C₂H₂) | Arcing | High-energy discharge, act quickly |
| CO / CO₂ | Paper and cellulose degradation | Insulation involvement |
A single high gas reading is rarely a diagnosis on its own. The pattern matters: high hydrogen with only small hydrocarbon levels points toward partial discharge, while acetylene presence, even at low parts-per-million levels, signals arcing and calls for a faster response than a purely thermal trend would.
Ratio methods and the Duval triangle turn gas combinations into fault types. They are not equally reliable: published comparisons put the key gas method around 42% and the Rogers ratio around 62%, while the Duval triangle does materially better — about 95% in Duval’s own comparison on the IEC TC 10 fault database, and typically 65–87% in independent studies. Multi-method or machine-learning fusion is reported between roughly 76% and 98% (research comparison figures, not product guarantees). Ratio methods need minimum gas concentrations to be valid, and a Duval zone is only meaningful when the relevant gases are clearly above normal.
None of this makes the tools useless. A good dga interpretation guide treats them as decision support, then confirms with trend direction, complementary diagnostics, and IEEE C57.104-2019 condition ratings (1-4) and their re-test intervals.
How fast does a dissolved hydrogen sensor respond to a developing fault?
For an online sensor, response speed is described by the T63 step response, set by the membrane permeation rate, oil cell volume, and oil flow. A fast-responding dissolved hydrogen sensor can flag a rise within minutes to hours, which is the difference between catching a fault between lab samples and finding it weeks later.
What hydrogen level in transformer oil should trigger action?
There is no single universal number; IEEE C57.104 gives percentile-based benchmarks and re-test intervals rather than one hard limit. What matters most is a sharp, sustained rise, so many operators act on rate of change first, treating a fast-climbing hydrogen trend as the trigger and confirming with a full gas profile. This dga interpretation guide applies the same logic to a lab report or a live online monitor.
PAS DGA for continuous dissolved hydrogen monitoring
Once the interpretation workflow tells you which transformers need attention, the next question is sampling frequency. PAS DGA online monitors, from single-gas dissolved hydrogen sensors such as the DGA-200, DGA-300 and DGA-500, up to the nine-gas plus moisture DGA-900, close the gap between quarterly oil samples. A continuous dissolved hydrogen trend supplies the rate-of-change data your workflow needs, and multi-gas coverage confirms the fault type when a hydrogen alarm fires. Contact PAS DGA to discuss which monitor fits your transformer fleet.