Acoustic Detection of Partial Discharge using Signal Processing and Pattern Recognition Techniques

Abstract

Power transformers are one of the major components in electric network. The service area of power transformers is quite large which means that the failure of such equipment will cause huge losses for power companies. Therefore, continuous monitoring of power transformers and preventing such failures is of great importance. During operation, power transformers are affected by different stresses such as electrical, thermal and mechanical stresses. Also, the presence of some defects within the transformer insulation under the applied electrical stress may initiate internal partial discharges. Previous experience has shown that internal partial discharges can cause serious damages for the insulation. Different methods like acoustic, optical and RF sensors were proposed to detect the partial discharge activity. Compared to other techniques, acoustic detection methods are cost effective and less susceptible to noise and interference. In addition, partial discharge (PD) source can be located using multiple acoustic sensors. However, acoustic methods have some drawbacks such as low sensitivity and high attenuation of acoustic signals. In this thesis, artificial neural networks are utilized to enhance the acoustic PD detection under different insulation conditions. Experimental results have shown that high detection rates of PD can be obtained under different conditions. Also, classification of insulation condition based on the acquired acoustic PD signals was successful.