Online DGA Monitoring: Fundamentals for Utility Asset Managers

Periodic laboratory DGA, collecting oil samples at scheduled intervals and analysing them by gas chromatography, has been the standard approach to transformer condition monitoring for more than five decades. It remains the foundation of any rigorous DGA programme. Its limitation is equally well understood: anything that happens between samples is invisible until the next one.

Online DGA monitoring addresses this gap with gas-in-oil sensors installed directly on the transformer, providing continuous or near-continuous concentration readings. For utility asset managers evaluating whether and where to deploy online monitoring, understanding the technology's physical basis, its genuine value proposition, and the analytical requirements for using it effectively is essential.

The Physical Basis of the Detection Advantage

The argument for online monitoring is not simply "more data is better." It is a specific claim about fault development timescales relative to sampling intervals.

Many serious transformer fault mechanisms develop on timescales that are shorter than a standard quarterly sampling programme. A thermal fault initiated by contact deterioration or incipient insulation failure can progress from initial gas generation to advanced condition within weeks. Electrical discharge, with partial discharge progressing to arcing, can intensify in days. For a transformer sampled quarterly, the window between the sample that first shows a developing fault and the sample that shows it has significantly advanced may be the entire 90-day interval.

For transmission-class transformers with long replacement lead times [1] and high operational consequence, that 90-day detection window represents a risk that is disproportionate to the cost of closing it with continuous monitoring. CIGRE TB 630 [2], produced specifically to address online monitoring for transformers, frames this lead-time argument as the primary justification for continuous monitoring deployment on high-consequence assets.

Sensor Technology and What Each Approach Measures

Commercial online DGA sensors use several analytical methods to measure dissolved gas concentrations [2]:

Photoacoustic spectroscopy (PAS) uses modulated infrared light to excite gas molecules extracted from the oil. The acoustic response to light absorption is proportional to concentration. PAS sensors can measure multiple gases simultaneously and are among the most widely deployed for multi-gas online monitoring.

Non-dispersive infrared (NDIR) measures light absorption at wavelengths specific to each gas. Also capable of multi-gas measurement, NDIR sensors are used by several commercial monitor manufacturers.

Electrochemical detection measures the electrochemical response of specific gas species at detection electrodes. Most commonly used for single-gas hydrogen monitoring due to its selectivity and low cost.

Gas chromatography (online GC) applies the same analytical method as laboratory analysis, automated for continuous or semi-continuous field operation. Online GC provides the most complete gas profile but at higher cost and complexity.

Single-gas vs. multi-gas. Hydrogen (H₂) is typically the first gas to appear in developing fault conditions and is detectable by the least expensive sensors. However, hydrogen alone does not distinguish between fault types: high H₂ with low hydrocarbons suggests partial discharge; H₂ accompanied by CH₄ and C₂H₄ suggests thermal fault; H₂ with C₂H₂ suggests electrical arcing. Multi-gas monitors that measure the full fault gas suite provide the fault type classification capability needed for a complete diagnostic response.

The Analytical Requirement: Why Threshold Alerts Are Insufficient

Raw online sensor data requires careful analytical handling before it produces useful results. This is the most commonly underestimated aspect of online monitoring deployment.

Online sensor readings carry noise sources that are not present in controlled laboratory extractions: temperature cycling drives apparent concentration changes as gas solubility in oil varies with temperature; load changes produce transient oil circulation effects; sensor drift accumulates between calibration intervals; and dissolved air variation can affect certain sensor technologies. Dukarm [3] established that analytical uncertainty is already substantial in laboratory DGA; online sensor readings can exhibit far larger short-term variation.

Applied directly to fixed threshold alerts, the approach that replicates conventional periodic sampling logic in a continuous format, raw sensor data produces frequent false alerts that erode operational confidence in the monitoring programme. This is the failure mode that prevents many online monitoring deployments from delivering their potential value.

The effective approach requires two elements:

Signal processing appropriate for continuous sensor data. Before any DGA calculation is applied, sensor readings should be conditioned to separate genuine gas concentration trends from noise sources. The specific signal processing required depends on the sensor architecture and the analytical method it uses.

Rate-of-change and trajectory-based alerting. The primary detection advantage of online monitoring over periodic sampling is sensitivity to fault initiation, the point at which gas concentrations begin rising above their baseline, well before any threshold is crossed. Detecting this requires rate-of-change alerting: flagging when the rate of gas accumulation has increased significantly relative to the recent baseline, regardless of the absolute concentration level. This is analogous to the HF (Hazard Factor) metric in R-DGA [4], which responds to changes in the current condition trajectory rather than to threshold exceedances.

Monitor Watch applies both elements: signal processing calibrated for online sensor data characteristics, followed by R-DGA calculation using the same CSEV and HF framework applied to laboratory samples in Transformer Oil Analyst™ (TOA) [4]. The result is a consistent risk picture for each transformer that treats laboratory and online data equivalently, enabling meaningful comparison and preventing the fragmented view that arises when two different analytical methods are applied to the same asset.

Selecting Transformers for Online Monitoring

Not every transformer in a fleet warrants the investment in continuous online monitoring. CIGRE TB 630 [2] recommends prioritising deployment on assets that combine high criticality (consequence of failure) with elevated condition risk indicators.

The practical selection criteria most commonly applied:

• High consequence of failure: Transformers at critical transmission substations, units serving load centres without backup supply, or units with replacement lead times that would create unacceptably long outages if they failed unexpectedly.
• Elevated fleet risk rank: Transformers in the upper HF quartile of the fleet, those whose R-DGA severity history indicates the highest population-relative failure probability [4]. Continuous monitoring on a transformer already identified as high-risk by fleet screening maximises the detection value of the programme.
• Identified fault trends: Transformers with existing DGA records showing consistent upward gas trends that have not yet crossed thresholds. These units have demonstrable fault activity underway and benefit from continuous visibility.

The combination of fleet screening through TOA and targeted online monitoring through Monitor Watch, one identifying which units warrant closer attention and the other providing it, creates the complete monitoring programme architecture that CIGRE TB 812 [1] and CIGRE TB 630 [2] together recommend for transmission asset management.

Practical Deployment Considerations

Data transmission and storage. Online sensor data should be integrated with the asset management and DGA analytics platform rather than stored locally at the sensor or reviewed manually. The analytical value of continuous data is fully realised only when it is processed automatically and compared systematically against the fleet baseline.

Calibration and maintenance. Online sensors require periodic calibration to maintain accuracy. Most sensor manufacturers specify calibration intervals; adhering to these is important for maintaining the data quality on which the analytical framework depends.

Integration with laboratory sampling. Online monitoring supplements rather than replaces periodic laboratory sampling. Laboratory gas chromatography provides validation of sensor readings and full gas suite analysis that most online sensors cannot match. Both data sources should feed into the same analytical framework.

For detailed information on Monitor Watch capabilities and integration options, visit the Monitor Watch page. For technical background on R-DGA methodology, visit the Science page. To discuss online monitoring deployment for your fleet, contact us.