The most recent power transformers implemented in the national transmission network are insulated with foreign insulating oils; these new insulating oils have a very high biodegradability and high flame and fire points. As these new oils feature better fireproof behavior, the ester insulated transformers are less exposed to the risk of fire and explosion and can also be pushed to operate at higher temperatures, which guarantees obvious advantages in terms of electrical system overloadability. However, the degradation mechanisms of the paper/ester insulating system, especially in particularly severe transformer operating conditions, are still little investigated and it is necessary to identify

diagnostic methodologies suitable for rapidly identifying any degradation processes. The research activity of the 2019-2021 three-year period concerns the analysis of the degradation problems of electrical power transformers insulated in foreign insulating oil and the related diagnostic methodologies.

The activity described in this report refers to the characterization phase of the MIP (Molecular Imprinted Polymer) interface by experimental verification of the optical response on 2-FAL standards in virgin natural ester. It was decided to focus on the detection of 2-FAL compared to other more recent indicators of paper degradation (such as methanol) for a more immediate comparison with what has already been acquired previously. The results obtained confirm that the optical response with MIP in ester is comparable to that obtained in mineral oil.

Some practical aspects concerning the configuration of the SPR (Surface Plasmon Resonance) platform with the synthetic receptor have also been explored. In particular, the effects of the variations of the optical sensor parameters and of the MIP thickness have been analyzed numerically and experimentally in order to optimize their performance and reproducibility.

During the research period, the experimental configuration needed to prepare laboratory tests to reproduce in a controlled manner different degradation conditions of the paper/oil system was also defined and test parameters (temperature, duration) were defined. Design solutions capable of overcoming oxidation issues related to the natural ester oil were identified, and particular attention was paid to the procedure for taking paper samples for comparison measurements with traditional methods (such procedure being particularly simplified). The main components of the test circuit were then identified and supplied.

As for the diagnostic methodologies for the identification of the transformer’s structural defects, the investigation continued to verify the possibilities of applying Machine Learning methods to real vibration data of the transformer under load for the classification of faults. The training of classifiers via SVM (Support Vector Machine) was carried out with data relating to a high number of measurement points (22) extended to different transformer tank positions to have a more complete classifier, capable of detecting any tightening loosening via measurements carried out in different parts of the tank (not limited to specific small areas). The obtained classifiers proved to be highly accurate for most of the positions used, and interesting indications emerged regarding the positions of the sensors considered more significant for the “training” of the system with a view to greater robustness of the classifier.