Aging Detection in Cast Resin Transformers by Vibration Data Analysis with Neural Networks

Vibration Detection is as an effective method to detect loose or deformed windings in transformers. Authors’ previous research work investigated the virtue of deep neural networks in detecting these faults in oil filled transformers. It was proven that neural networks can detect loosening fault with a high accuracy, robustly to possible misplacements in the positioning of vibration sensors, under load and no-load transformer’s operation. In this paper, the analysis of vibrational spectra through neural networks is applied to a cast resin transformer, with the aim of evaluating whether progressive changes in the sampled vibrational pattern can be correlated with the aging of the transformer insulation. The proposed approach yields a classifier capable of predicting the transformer aging with satisfactory accuracy and suggests a linear and progressive deterioration of the transformer over time.