Understanding CSEV and Hazard Factor: The Metrics Behind R-DGA

Reliability-based DGA (R-DGA) produces two primary outputs: Cumulative Severity (CSEV) and Hazard Factor (HF). These are not proprietary repackagings of conventional DGA metrics. They represent a methodologically distinct approach to transformer risk assessment, grounded in statistical analysis of a large population of transformer failure histories rather than in fixed concentration thresholds.

Understanding what CSEV and HF actually measure, and why the population-based approach they embody is more informative than threshold interpretation, is essential for asset managers evaluating whether to adopt R-DGA methodology.

The Fundamental Limitation of Threshold-Based DGA

The standard approach to DGA interpretation, codified in IEEE C57.104-2019 [1] and IEC 60599:2022 [2], compares individual gas concentrations against published limits. Hydrogen (H₂) above a certain ppm level triggers a Condition 2 or 3 classification; acetylene (C₂H₂) above its limit triggers immediate follow-up; total dissolved combustible gas (TDCG) above a threshold value indicates progressive concern.

This approach has a fundamental structural problem: it evaluates each DGA sample in isolation, at a single point in time, against limits that do not account for the transformer's individual history or for the statistical distribution of gas concentrations in actual transformer populations.

Consider two transformers: Transformer A has hydrogen at 200 ppm, which it reached over 15 years of stable operation at consistent loading: a gradual, long-term profile that places it in the middle of the normal population distribution for transformers of its age and type. Transformer B also has hydrogen at 200 ppm, reached after a 12-month period during which concentration tripled: a trajectory indicating active fault development. Against IEEE C57.104 [1], both transformers are evaluated identically: both sit in the same condition level, both trigger the same follow-up recommendations. The rate of change, the trajectory, and the population context are invisible to the method.

Dukarm et al. [3] demonstrated systematically that this invisibility produces both false alarms, flagging transformers whose gas levels are elevated but entirely normal for their population, and missed detections, failing to identify transformers whose trajectory indicates genuine risk accumulation that has not yet crossed a threshold. Both failure modes carry costs: unnecessary maintenance expenditure for false alarms, and unexpected failures for missed detections.

Cumulative Severity (CSEV): A History-Integrated, Population-Normalised Measure

CSEV addresses the first limitation, isolation of each sample, by integrating over the transformer's full DGA history.

The method was developed by Jim Dukarm, Ph.D. [4] and refined over three decades of validation against utility transformer population data. The calculation weights dissolved gas measurements by their diagnostic significance, with hydrogen and ethylene indicating thermal and electrical fault activity, acetylene indicating high-energy electrical events, and CO and CO₂ indicating cellulose insulation degradation, and accumulates a fault severity score across the transformer's historical record.

This accumulated score is then normalised against the reference population of transformer histories. The normalization is what gives CSEV its meaning: a CSEV of 70 means this transformer's cumulative gas profile is more severe than 70% of the reference population. It sits at the 70th percentile of fault severity for transformers with comparable historical records.

Several properties of CSEV are worth emphasising:

It is trajectory-sensitive. A transformer that has generated fault gases steadily for 20 years will have a different CSEV than one that generated the same total amount of gas in two years. The rate of accumulation affects the profile of the integral.

It is noise-resistant. Single DGA measurements carry analytical uncertainty that can be substantial: Dukarm [5] demonstrated coefficients of variation in the 10–20% range for most dissolved gases under standard laboratory conditions. A single erroneous high reading inflates a concentration-based alert immediately, but has a proportionally small effect on a CSEV value built from dozens of historical samples.

It is comparable across a fleet. Because CSEV is normalised against the reference population, it produces a number with the same meaning for every transformer in the fleet, regardless of voltage class, age, oil volume, or design. This makes it directly suitable for fleet ranking.

Hazard Factor (HF): Population-Calibrated Failure Probability

CSEV tells you where a transformer sits in the population distribution of cumulative fault severity. HF tells you what that position means for failure probability.

HF is derived from the empirical relationship between CSEV level and observed transformer failure probability in the population reference data [3]. The mathematical framework is borrowed from reliability engineering: the hazard function describes the instantaneous failure rate as a function of cumulative damage accumulated. In the transformer DGA context, CSEV serves as the damage proxy, and the hazard function is calibrated against the distribution of CSEV values at which failures actually occurred in the reference population.

A transformer with HF of 0.8 has a cumulative severity profile that, in the reference population, corresponds to a failure rate 0.8 times the rate seen at the median failure event in the data. This is not a precise predictive probability. No DGA metric provides that. It is, however, a defensible, empirically calibrated statement of relative risk that places the transformer's condition in the context of what the data says about transformers at comparable CSEV levels.

HF is more sensitive to recent changes in the gas generation trajectory than CSEV. A transformer whose gas generation rate accelerates recently will show CSEV rising slowly (because the history is long) but HF rising more quickly (because the hazard function responds to the current CSEV level relative to the failure distribution). This makes HF the metric most useful for identifying transformers requiring immediate attention, while CSEV provides the longer-term severity context.

How CSEV and HF Complement Fault Classification Methods

The Duval Triangle [6] and the IEC 60599 ratio methods [2] are designed to classify the type of fault indicated by a transformer's gas profile: low-temperature thermal fault, high-temperature thermal fault, partial discharge, arcing. This classification is useful diagnostic information, informing what kind of inspection or test to perform next, but it does not address severity or urgency.

R-DGA is not a replacement for fault type classification. CSEV and HF add a quantitative severity dimension that fault classification methods do not provide. A transformer with a high-temperature thermal fault classification and HF of 0.3 may be monitored on an accelerated schedule. The same classification at HF of 0.9 warrants immediate investigation. The fault type tells you what to look for; CSEV and HF tell you how urgently to look.

Transformer Oil Analyst™ (TOA) presents both: fault type classification consistent with standard methods alongside R-DGA severity metrics, so asset managers have the full picture.

Practical Impact on Fleet Management

In fleet management practice, CSEV and HF together produce a ranked list of transformers ordered by population-normalised risk, updated automatically each time new DGA results are imported. Asset managers can sort by HF to identify units requiring immediate attention, by CSEV to identify units with the most severe cumulative histories, or by a combination to identify the segment of the fleet where both metrics are elevated simultaneously.

Utilities that have adopted this approach consistently report the same two findings: a subset of transformers that conventional methods were flagging have CSEV/HF profiles within the normal population range, false alarms that can be de-escalated, and a different subset that were not triggering threshold flags but whose CSEV trajectory places them in a population segment with elevated failure probability [3].

Both findings have direct financial value: reduced unnecessary maintenance expenditure for the first group, and advanced warning enabling planned rather than reactive intervention for the second.

For the technical basis of R-DGA methodology including the full mathematical formulation of CSEV and HF, visit the Science page. For product information on TOA, visit the TOA page.