Abstract
The Frequency Response Analysis (FRA) is an advanced method for diagnosis of failures in the active part of power transformers. The assessment of FRA results relies on the comparison of a reference FRA curve to an actual curve and based on the deviations between the two curves and the experience of a human expert, an assessment about the integrity of the components of the active part is issued. At the present there is a lack of reliable algorithms for automatic assessment of the results. This motivated to research new methodologies for overcoming this problem. As a contribution to this necessity, this paper summarizes the outcome of a research work in which machine learning algorithms were used for automatic assessment of FRA measurements. Decision tree classifiers were developed using the algorithm C4.5. The results obtained give evidence of the effectiveness of the proposed classifiers.
Application of machine learning techniques for automatic assessment of FRA measurements
