Choosing a DGA Interpretation Method
Once you have a set of dissolved gas readings, the next question is how to turn numbers into a fault verdict. The industry has developed several families of methods — single-component thresholds, ratio codes, graphical zoning and, more recently, machine-learning fusion. They are not interchangeable: each has a different principle, different data requirements and different blind spots.
This guide compares the methods you will actually meet in practice, summarizes what independent comparative studies say about them, and recommends a workflow rather than a single winner.
The Key Gas Method: Simple but Qualitative
The key gas method judges a fault from single-component concentrations and total dissolved combustible gas (TDCG). Its appeal is simplicity — you compare a few numbers against thresholds and get a fast qualitative answer in the field. Its limitation is equally clear: it is qualitative only, with low resolution for combined faults. A single gas cannot separate a discharge from an overheated connection, which is why the method is best used as a first screen, not a final diagnosis.
Ratio Methods: Doernenburg, IEC and Rogers
Ratio methods encode the relative proportions of characteristic gases, which removes much of the dilution effect of oil volume and degassing history.
- Doernenburg ratio method — uses four gas ratios and is effective for specific fault scenarios, but many combinations simply cannot be classified.
- IEC gas ratio method — codes three characteristic-gas ratios into discrete codes. It is quantitative and automatable, but combinations that fall outside the predefined codes cannot be diagnosed.
- Rogers ratio method — the classic three-ratio coding scheme, widely used and familiar to most DGA practitioners, but with incomplete combination coverage and a tendency to misclassify certain scenarios.
All three share one structural weakness: when a real fault produces a gas combination that is not in the code table, the method goes silent or, worse, gives a confident wrong answer.
Graphical Methods: Duval Triangle and Pentagon
The Duval triangle normalizes the three principal hydrocarbon gases (CH4, C2H2, C2H4) and locates the fault in a triangular zone, with separate triangles for different operating conditions. It is simple to implement and graphically intuitive, but it has limited resolution for mixed faults and for low-temperature thermal faults.
The Duval pentagon extends the idea to seven zones (PD*, D1*, D2*, T3*, H*, C*, O*, S*) and requires full-component data. In research comparisons it shows the best consistency of any single method, largely because it adds resolution where the triangle is weakest — low-temperature thermal faults and the carbonization region.
Accuracy: A Research Comparison
The figures below come from laboratory comparative studies and should be read as a research comparison of relative merits, not as an absolute or product-accuracy promise. Results vary across study samples and datasets.
| Method | Principle | Strength | Limitation | Accuracy (research comparison) |
|---|---|---|---|---|
| Key gas method | Single-component concentrations + TDCG | Simple, intuitive, field-friendly | Qualitative only; low resolution for combined faults | ~42% |
| Doernenburg ratio | Four gas ratios | Effective for specific scenarios | Many combinations unclassifiable | — (insufficient samples) |
| IEC gas ratio | Coding of three ratios | Quantitative, automatable | Codes outside the table cannot be diagnosed | Moderate |
| Rogers ratio | Coding of three ratios | Classic, widely used | Incomplete coverage; some misclassification | ~62% |
| Duval triangle | Zoning by CH4/C2H2/C2H4 | Graphically intuitive, easy | Limited for mixed and low-temperature faults | ~95% |
| Duval pentagon | Seven zones (PD*/D1*/D2*/T3*H*/C*/O*/S*) | Best consistency of single methods | Requires full-component data | Comparable to the triangle (dataset-dependent) |
| Multi-method / ML fusion | Voting or machine-learning fusion | Integrates criteria, high robustness | Depends on sample quality and interpretability | 76–98% (varies by study) |
Figure sources: the Duval triangle value is from M. Duval’s method comparison on 122 cases from the IEC TC 10 fault database (reported at 95–96% for Duval triangles I and II); the key-gas figure comes from the same IEC TC 10 fault-database comparisons, where only about 42% of key-gas diagnoses were correct. The ratio-method limitation is structural rather than statistical: the IEC ratio scheme defines only 11 of its 27 possible code combinations, so any reading that falls outside the table returns no diagnosis at all. Independent academic benchmarks report a wide spread — roughly 65–87% for the Duval triangle — depending on dataset and fault mix. All values are fault-classification hit rates on research datasets: they are not measurement accuracy, and they are not product specifications.
Read the numbers carefully — and do not read them as a league table. Two things are consistent across the literature: the key gas method is the weakest screen, and the ratio methods return nothing at all when a gas combination falls outside their code table. The graphical methods do materially better. Machine-learning fusion is reported from roughly 76% to 98%, but on small or unbalanced samples its apparent edge over a well-applied Duval triangle can disappear. The honest argument for combining methods is not that one number beats another — it is that the different methods fail in different places.
Multi-Method Voting, ML Fusion and a Practical Workflow
Because each method has different blind spots, running several in parallel and combining their verdicts is far more robust than relying on any one. Simple voting already improves consistency; machine-learning fusion goes further by learning which methods to trust for which gas patterns. The practical caveats are data quality and interpretability — a fused verdict needs enough labeled samples behind it, and an engineer needs to be able to explain the outcome.
For a step-by-step approach to running multiple methods on the same data set, see our multi-method DGA diagnosis workflow.
The defensible approach in the field is a three-layer ladder. First, screen with the key gas method and TDCG to flag any anomaly. Second, run the IEC ratio and Duval methods on the flagged readings — the pentagon when full-component data is available. Third, when methods disagree or a critical asset is involved, bring an expert into the loop and treat any machine-learning output as a decision-support input, not an automatic verdict.
What all of this depends on is complete, reliable multi-gas data. You cannot run Duval methods, ratios or fusion on a monitor that only reports hydrogen.
PAS DGA for Multi-Method-Ready Data
The PAS DGA DGA-900 online monitor supplies the full nine-gas-plus-moisture data set that every method in this article expects — including acetylene down to ≤0.1 ppm. That single feed keeps the key-gas screen, IEC ratios, Duval triangle and pentagon, and any fusion layer all running on consistent, comparable data.
Browse the diagnostic methods hub for more detail, or contact PAS DGA to discuss a multi-method monitoring setup for your assets.