Dissolved gas analysis (DGA), although not an exact science, has been used for decades to assess the condition of transformers because of its ability to detect and diagnose internal faults. On-line monitoring technology was introduced in the late 1990s, and today there are many on-line gas monitoring systems to choose from. The difficulty is that each detector has its own technical specifications, which makes it difficult for transformer owners to compare and evaluate among multiple options. In addition, laboratory DGA makes the field more complex.

Tip: Click the link at the end of this blog post to download the white paper.

Measurement accuracy

Inaccurate DGA results can lead to faulty fault diagnosis, especially if the gas ratio is close to the boundary of the fault area. In addition, if the concentration value is close to the alarm value used by the facility, inaccurate results may cause the wrong measures to be taken against the transformer. Therefore, understanding uncertainty and measuring performance is crucial.

Measurement performance is defined by dynamic characteristics such as measurement range, response time, sensitivity, accuracy, and stability (referring to the instrument’s tolerance to aging and harsh environments). Among them, accuracy is often considered the most important and one of the most difficult to specify, which may or may not include repeatability, which refers to the ability to provide similar results when repeated measurements are made under constant conditions. However, it may not include long-term stability. Repeatability alone is often a secondary source of measurement uncertainty, and if accuracy does not include other uncertainties, actual measurement performance in real-world applications may give a false impression.

PW702

Laboratory DGA

Laboratory DGA is influenced by a variety of factors, from the quality of the oil sample to the equipment to the criteria used for analysis, not to mention the human element that is always present when manual processes are used. The most common sources of uncertainty include oil sampling methods and quality, gas extraction methods, gas distribution coefficients used, different criteria used, etc. It is also important to understand that the measurement results cannot be more accurate than the reference used in calibration.

The biggest source of uncertainty is usually sample quality. Large amounts of gases such as H2 and CO may escape from the oil, or ambient gases in the air, such as oxygen and nitrogen, may contaminate the sample, all of which can lead to errors in laboratory analysis. Therefore, the oil must not come into contact with air at any time during the sample collection process, and the sample container must be completely filled. In order to achieve this, the best way is to use high-quality syringes or aluminum bottles, as they can withstand factors such as pressure changes during air freight.

The IEC 60567 standard recommends that each laboratory determine its own accuracy or uncertainty and make this information available to its users, which is also a requirement for accredited laboratories to comply with. If there are no official figures for uncertainty, ask if the laboratory participates in any international inter-laboratory comparison tests (known as loop tests (RRT)) and if their results are available. This is a good indication of the approximate level of uncertainty. For laboratory DGA, there are two commonly used standards worldwide: IEC 60567 and ASTM D3612. It should be noted that ASTM and IEC standards calculate gas volumes at different temperatures at 0°C and 20°C, respectively. This alone brings about an 8% difference in the defined concentration of the same sample, which must be taken into account when comparing DGA results from monitoring systems and laboratories. All measured ppm values should first be converted to the same conditions, i.e. 20°C (IEC) or 0°C (ASTM).

On-line DGA monitoring system

The online DGA monitoring system, which measures seven critical fault gases, can identify multiple types of internal transformer failures at an early stage that might not have been detected between conventional oil sampling intervals without this method. In laboratory analysis, in order to obtain useful inputs for transformer condition assessment, each oil sample and its analysis must be representative. On-line monitoring systems provide greater flexibility and can also ensure the reliability of the data used for diagnosis by averaging it. Unaveraged data can be used to quickly diagnose evolving faults. Using an online monitoring system to track the rate of gas change is more reliable than using laboratory samples.

Most monitoring systems specify their accuracy for a traceable reference gas, but some monitoring systems may use a standard sample of gas in oil as a reference. The DGA monitoring system delivered should always be accompanied by a calibration certificate showing the difference between the monitoring system and the reference. In addition, the calibration certificate should specify the reference method used and whether the calibration can be traced to an international reference. However, the reported accuracy specifications are not entirely directly applicable to actual transformers in operation, as the oil in the transformer and its distribution factor are likely to be different from the oil and its distribution factor used in the calibration of the monitoring system. The ideal way to truly understand the performance of a monitoring system is to test it for a longer period of time, such as six months, using a live transformer. At least three to five oil samples should be taken at the same time, preferably by two independent laboratories that can provide uncertainty values for their own processes.

Compare on-line monitoring system and laboratory DGA

When evaluating an online monitoring system by comparing it to laboratory reference data, the uncertainty of sample quality and laboratory processes must be taken into account. It is also important to keep in mind that any analysis method, whether it is a laboratory or an online monitoring system, has its own uncertainties. The above factors should therefore be taken into account when comparing results and drawing conclusions about monitoring performance. Also note that even if the sample is perfect, there will be some variation in the results, and there may be significant biases if the methods used follow different criteria.