There is a version of transformer monitoring discussions where DGA and partial discharge detection end up positioned as competitors for the same budget line. That framing misses what each method actually does. They detect different physical phenomena, they have different detection lead times, and the fault categories they cover have only partial overlap. A monitoring strategy that understands this uses them in combination rather than as substitutes.
This piece covers what each method catches, where each one has blind spots, and how EM signature analysis fits into the picture as a third input covering different early fault behavior.
Dissolved Gas Analysis: What It Measures and When
DGA measures gases dissolved in transformer insulating oil. Under electrical and thermal stress, the mineral oil and cellulose paper insulation decompose and produce characteristic gases: hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide. The pattern of gases and their ratios, interpreted against the key gas method in IEEE C57.104 or the IEC 60599 ratio method, gives a fingerprint of the fault type and approximate severity.
The four critical gases for fault type discrimination are hydrogen, acetylene, ethylene, and CO. Hydrogen is produced by partial discharge and low-temperature oil decomposition. Acetylene appears under high-energy arcing. Ethylene is a thermal decomposition product, dominant in hot-spot faults above roughly 700C. CO and CO2 track paper insulation involvement.
What DGA does very well: it characterizes the physical chemistry of the fault state. A transformer with 180 ppm acetylene and a hydrogen-to-methane ratio consistent with arcing is exhibiting a specific failure mode that can be interpreted with considerable confidence. The Duval triangle and Rogers ratio methods exist precisely because this gas chemistry interpretation has been validated across thousands of transformer failure histories. The correlation between specific gas patterns and specific fault families is one of the better-established relationships in transformer diagnostics.
What DGA does not do well: early detection of developing faults. The gas must dissolve into the oil, reach a measurable concentration, and be captured in a sample before DGA can register it. Dissolved gas equilibration between the fault site and the bulk oil takes time. For an intermittent partial discharge source in the winding, the hydrogen generation rate may be too low to produce a concentration change detectable above the measurement uncertainty of the lab analysis for weeks or months. A single annual sample at below-threshold values can pass a transformer that has an active, developing fault.
Online DGA monitoring addresses the sampling interval problem, but not the sensitivity threshold or the equilibration lag. Gas generation rate trending over time (rather than absolute concentration) improves detection speed for slowly developing faults, but the fundamental sensitivity limitation remains.
Partial Discharge Detection: Method and Coverage
Partial discharge (PD) is a localized electrical discharge within the insulation system that does not fully bridge the gap between conductors. PD occurs in voids in the solid insulation, at contamination sites in the oil, at conductor edges with insufficient clearance, and in the interface regions between paper and oil in the winding structure. It produces a brief current pulse with associated electromagnetic emissions across a wide frequency range, acoustic emissions from the mechanical shock of the discharge event, and chemical byproducts that eventually appear in DGA.
Electrical PD measurement on in-service transformers typically uses high-frequency current transformers (HFCT) at bushing connections or neutral grounds to capture the PD pulse signatures. Acoustic monitoring using piezoelectric sensors on the tank wall triangulates discharge source location. Both methods detect active PD in near real-time, which gives them a detection advantage over DGA for faults where PD activity precedes significant gas production.
PD detection is highly sensitive for certain fault types. Voids in solid insulation, contaminated oil under high electrical stress, and deteriorating bushing paper are all strong partial discharge sources. For these, a continuous PD monitoring system will detect fault onset well before DGA shows elevated hydrogen, in some cases with detection lead times of weeks to months ahead of a significant gas result.
The limitation of PD detection is coverage and noise. Not every fault type involves significant PD activity. A winding with thermally degraded insulation approaching the end of its dielectric life may have low PD levels until the final breakdown event. A hot-spot developing in the winding from a turn-to-turn circulating current may generate relatively little PD while producing significant thermal decomposition products. The HFCT measurement is also susceptible to interference from external switching events, cable discharge from adjacent equipment, and ground loop noise in substation environments, which requires careful installation and signal filtering to avoid false indications.
Where the Two Methods Diverge by Fault Type
Setting this out clearly is more useful than abstract discussion:
Partial discharge in void / contaminated oil: PD detection catches this early. DGA will show hydrogen elevation weeks to months later.
Bushing insulation deterioration: PD detection is excellent if the bushing has an active discharge. Bushing power factor and capacitance measurements (offline or online) are the gold standard. DGA will reflect bushing paper involvement eventually but is a lagging indicator.
Hot-spot from overloading or circulating current: DGA (ethylene and methane trend) is the primary indicator. PD may be minimal until the fault produces a discharge-prone geometry. Thermal imaging can supplement but requires accessible hotspot location.
Turn-to-turn winding fault development: This is the gap where both methods have limitations. Low-level circulating current through a partially shorted winding turn may produce modest gas generation and modest PD levels in the early weeks. The fault does produce a change in the current distribution within the winding, which appears in the harmonic content of the transformer's external electromagnetic output before it produces gas or discharge at levels reliably detectable by the other methods.
Core fault: DGA (CO, CO2, hydrogen) detects core insulation involvement. Core vibration sensors directly. PD levels are variable. EM signature is affected as core saturation characteristics change.
The Electromagnetic Signature Layer
We designed Magnefy's monitoring approach around a specific observation from field experience: the early phase of a winding fault produces an EM signature change before it produces a DGA-detectable gas change or a reliably measurable PD signal. This is not a surprising finding from first principles. Any change in the effective winding impedance or current distribution alters the harmonic composition of the transformer's output field. A partially shorted turn creates a low-impedance parallel path for a portion of the winding current, changing the relative magnitude and phase of the harmonic components. The change is small in absolute terms, but it is systematic and continuous from fault onset.
The practical challenge is separating this signal from the background variation caused by load changes, temperature, and ambient electrical noise. The load-induced variation in EM harmonic content can be 10 to 20 times the magnitude of the fault-induced change in the early stages. This is why the method requires a calibrated baseline and load-normalized tracking, not just a fixed threshold comparison.
We are not claiming that EM signature analysis replaces DGA or PD detection. What we are saying is that it covers a detection window for specific fault families, particularly winding insulation onset and early turn-to-turn fault development, that the other two methods do not reliably catch at the same lead time. A utility running annual DGA plus a continuous EM monitor has a combination that catches most of the major fault families with reasonable lead time: the EM monitor identifies early winding fault onset and triggers a targeted DGA sample plus field inspection, while the DGA provides the chemistry characterization that supports the maintenance decision.
Practical Integration of the Three Methods
For a distribution fleet of 150 to 400 transformers with a mix of ages and criticalities, the monitoring architecture that makes economic sense typically looks like this: continuous EM monitoring on all units in the fleet (or the subset above a criticality threshold), online DGA on a smaller set of high-criticality units such as main substation transformers and step-up transformers at industrial facilities, and periodic offline DGA on the remainder, with sampling interval driven by fleet age and historical fault rate rather than a fixed schedule for all units equally.
The continuous EM monitor acts as the fleet-level triage layer: it identifies the units that need additional attention. Those units receive accelerated DGA sampling, and if the DGA confirms gas evolution trends, a PD measurement and field inspection are scheduled. This concentrates the higher-cost diagnostics on the units that have shown early warning behavior, rather than distributing them uniformly across a fleet where most transformers are behaving normally at any given time.
It is worth being direct about the limitation of this picture: the EM method's sensitivity for early winding fault detection has been validated on a modest number of units in controlled monitoring conditions. It is not at the same level of documented field validation as DGA, which has decades of standardized use behind it. We are careful about the claims we make. What we have observed consistently in early-access pilot programs is EM anomaly lead times that precede DGA threshold crossings by 3 to 6 weeks for developing winding faults. That is useful, but it is early-stage data, not a large sample from diverse fleet conditions.
The honest answer is that transformer monitoring is not a solved problem. Each method covers a part of the fault space. Understanding what each method catches, and where each one is silent, is the prerequisite for building a monitoring program that actually reduces the rate of unplanned outages rather than simply generating more data.