Enhancement of acoustic based partial discharge detection using pattern recognition techniques

Partial Discharge (PD) phenomenon is one of the major factors that may lead to insulation deterioration in power transformers. Several techniques were developed to detect partial discharge activity. Acoustic detection has been utilized for PD signal detection in power transformers. Acoustic detection has several advantages compared to other techniques such as: it is immune to electromagnetic interference and it can be used to locate the PD activity. However, the acoustic signals suffer from high attenuation which makes the detection of PD activity a difficult task. This paper presents a pattern recognition based technique for enhancing the acoustic detection of partial discharge signals. Different cases for PD generation were simulated which include the presence of different types of barriers such as paper insulation and core material. In addition, the effects of the tank size and the distance between the PD source and the acoustic sensor on the detection performance were studied. The features extracted from the acquired signals in all cases were fed to an artificial neural network which was used for training and classification. The results have shown that the detection performance of acoustic PD signals could be significantly enhanced using some features like signal entropy.