The most recent power transformers integrated into the national electrical grid are insulated with ester oil. Ester oils have a very high biodegradability and a high flash point, which makes ester-insulated transformers less prone to fire risk and capable of operating even under overload conditions. However, to date, the factors triggering accelerated degradation of the paper-ester insulation system, especially during particularly demanding transformer operation, are still not well understood. Accelerated degradation of the paper-ester system can affect the expected lifespan of a transformer and can also negatively impact the tightening of the coils. Therefore, it is necessary to identify diagnostic methodologies that can detect indicators of this process at an early stage.
In this report, the possibility of extending the application of dielectric and opto-chemical diagnostic methodologies, previously tested in mineral oil, to the new insulating oil matrix (ester) has been positively verified. Specifically, the dielectric behavior at different frequencies of a new paper-ester insulation system for two different types of paper (traditional Kraft and thermally upgraded Kraft, TUK, both dried in VP, Vapor Phase) and virgin natural ester was analyzed. Compared to mineral oil, it was confirmed that the higher affinity of esters for water results in an increase in the dissipation factor of the solid insulation system, especially at low frequencies. On the other hand, with the same insulating oil, the dielectric response does not seem to depend on the type of paper but on the drying process of the paper.
Regarding the opto-chemical methodology, the optical response on furfural (2-FAL) standards in samples of two families of natural esters was analyzed, which are comparable to that obtained in mineral oil. The effects of the thickness of the sensitive polymer layer made with molecular imprinting technology (Molecular Imprinted Polymer, MIP) and of an extended washing of the same on the quality of the analytical signal were also analyzed. The spectra are more defined if the MIP layer is thinner, while a lower detection limit of the sensor is obtained after extended washing. The latter could improve the detection of 2-FAL, particularly in the case of TUK paper degradation.
The good selectivity of the MIP for furfural (2-FAL) compared to another furanic compound, 5-hydroxymethylfurfural (5-HMF), is confirmed, despite the considerable similarity between the two molecules. In view of subsequent validation of the aforementioned methodologies on samples under known and controlled degradation conditions, the experimental circuit was assembled, which simply and originally reproduces the operating conditions of a sealed ester-insulated transformer. To increase the reliability of fault diagnosis (coil loosening) of a transformer, classification methods based on neural networks (instead of Support Vector Machines) were chosen for the analysis of vibration data, in order to consolidate the calibration procedure of a robust classifier. The new code also preliminarily addressed a second and new case study involving resin transformers subjected to thermal stress. The results seem to confirm that, even for these transformers, vibration data provide information about the aging state of the transformer.