Although it is not an exact science, dissolved gas analysis (DGA) has been used to assess transformer condition for decades, as it is the only method that can detect and diagnose internal faults in transformers. Online monitoring has been available since the late 1990s and today there are various online gas monitors available. The challenge is each one has its own technical specification, which makes it challenging for transformer owners to compare and evaluate the different options. To make the field even more complicated, laboratory DGA also plays its own role.
TIP: Download a related white paper via the link at the end of this blog post.
Measurement accuracy
Inaccurate DGA results can lead to faults being misdiagnosed, especially if gas ratios are close to a fault zone boundary. In addition, inaccurate results can cause the wrong action to be taken on a transformer if concentration values are close to the alarm values used at a utility. This makes it essential to understand uncertainty and measurement performance.
Measurement performance is defined by dynamics such as measurement range, response time, sensitivity, accuracy, and stability, meaning tolerance for aging and harsh environments. Of these, accuracy is often considered the most important quality. It is also one of the most difficult to specify; it may or may not include repeatability, which is the capability to provide a similar result when the measurement is repeated under constant conditions. However, it probably does not include long-term stability. Repeatability alone is often a minor source of measurement uncertainty, and if the accuracy specification does not include other uncertainties, it may give the wrong impression of actual measurement performance in real application.
Laboratory DGA
Laboratory DGA is affected by many factors starting from the quality of the oil sample to the equipment and standard used for analysis, not to mention the human factor, which is always present when manual processes are applied. The most common uncertainty sources include the oil-sampling method and quality, the gas extraction method, the gas partition coefficients used, the different standards used, and so forth. It is also important to understand that a measurement cannot be more accurate than the reference used in calibration.
The biggest uncertainty source is usually sample quality. A significant quantity of gases such as H2 and CO can escape from the oil, or ambient gases in the air, such as oxygen and nitrogen, can contaminate the sample – all of which will result in erroneous analysis in the laboratory. It is therefore essential that the oil is not in contact with the air at any point during sample collection, and the sample vessel must be completely filled. The best way to ensure this is to use high-quality syringes or aluminum bottles, which can, for example, tolerate pressure variations during air cargo transportation.
The IEC 60567 standard recommends that each laboratory should determine its own accuracy or uncertainty and make this information available for its users – and this is a requirement for accredited laboratories. If official numbers for uncertainty are not available, it is worth asking if the laboratory has participated in any international inter-laboratory comparison tests, known as round robin tests (RRT), and if the results are available. This gives a good indication of the approximate uncertainty level. Two standards are commonly used globally in laboratory DGA: IEC 60567 and ASTM D3612. It is important to note that ASTM and IEC standards calculate gas volume at different temperatures, 0°C and 20°C respectively. This alone brings an ~8% difference to defined concentrations for identical samples, which has to be taken into account when comparing DGA results from a monitor and a laboratory. All measured ppm values should first be converted to the same condition, either 20°C (IEC) or 0°C (ASTM), depending on preference.
Online DGA monitors
Online DGA monitors, which measure all 7 key fault gases, can identify all types of internal transformer faults at an early stage, when they might otherwise go unnoticed between regular oil sampling intervals. In laboratory analysis, in order to get useful input for a transformer condition assessment, every oil sample and their analysis must be representative. With online monitors, there is more flexibility and averaging can also be used to ensure reliable data for diagnostics. Without averaging, data can be used to quickly diagnose an evolving fault. Tracking the rate of change of gases with online monitoring is more reliable than with laboratory samples.
Most monitors have their accuracy specified at the point of calibration against traceable reference gases, while some may use gas-in-oil standard as reference. A delivered DGA monitor should always be accompanied by a calibration certificate showing the difference between the monitor and the reference. Additionally, it should specify the reference method used and whether the calibration is traceable to international references or not. But reported accuracy specifications are not applicable directly to a real transformer in operation, because the oil in a transformer and its partition coefficients are most likely not the same as the ones used in the monitor calibration. The best way to get a true picture of a monitor’s performance is to test it over a longer period, e.g. six months with a live transformer. Simultaneously at least three to five oil samples should be drawn, preferably for two independent laboratories who are able to supply the uncertainty values for their own processes.
Comparing online monitor and laboratory DGA
When evaluating an online monitor by comparing it to lab references, the quality of the samples and the uncertainty of laboratory procedures must be taken into account. Furthermore, it is important to remember that every analysis method – whether lab or online monitor – has its own uncertainties. These should be considered when comparing results and making conclusions on monitor performance. It must also be remembered that there will be some differences in results even if the sample is perfect, and significant deviation is possible if the methods used follow different standards.
