Clasificación de patrones de descargas parciales en aceites vegetales mediante Clustering DBSCAN

Abstract

This research characterizes and diagnoses partial discharge patterns in vegetable oils under different contamination levels. Through a quantitative experimental methodology, a high-voltage test bench was used to simulate controlled internal defects. The signals, acquired with a digital oscilloscope, underwent a filtering preprocessing stage to eliminate noise. Subsequently, the DBSCAN clustering technique was applied to process the information. The results were validated by comparing the response of the virgin oil against samples contaminated with moisture and conductive particles. It was demonstrated that the DBSCAN algorithm effectively discriminates pulse morphology and isolates the specific density cores for each evaluated contaminant.