Analysis of DC low energy discharges emitted light for transformer oil state assessment

The present article covers simultaneous measurements of current, emitted light and voltage drop associated with DC discharges in power transformer mineral oil. These measurements are analyzed in both time and frequency domains. The study focuses on the nanosecond scale for electrical signals and on the megahertz scale for spectral analysis. All of these measurements are characterized by high-frequency activity between 20 and 125 MHz and are highly correlated. By using fast response sensors, the signals have been acquired and represented in a digital format. This representation has been used to differentiate between oil samples with different water content based on FFT light signals analysis. For in service aged oils, it was difficult to differentiate between their aging when based only on the latter approach. Then, a feature extraction of light emission signals and their classification based on SVM have been successfully performed. In the experimental section, it has been shown that the proposed approach is capable of distinguishing between two different classes of oil (new or aged). Hence, by the use of simple SVM protocol, aged oils are well identified with an accuracy of 99% by analyzing light signals provided from low energy discharges. Such techniques can be useful for sealed power equipments diagnosis.