Condition Monitoring of Oil-filled Transformers using Unsupervised Classification Techniques
Résumé
In this contribution, an automatic classification technique is used to monitor oil-filled transformers condition from maintenance data. The data collected from two groups of transformers are reported into an observation space. This space summarizes the information contained in a feature vector. This vector is filled with oil’s aging indicator parameters namely the dielectric strength, the interferential tension, acidity and power factor. The TDCG (Total Dissolved Combustible Gas) is included to monitor incipient failure in the transformer. Two unsupervised classification methods are used: the K-means and Fuzzy C-means. The clusters included transformers with similar behavior to ease analyses.