كاربرد تشخيص الگوي تخليه جزئي در ترانسفورماتورهاي مقياس كوچك داراي نقص عايقي

Applicationof Partial Discharge Pattern Recognition on Small Scale DefectedTransformers

Partial Discharge (PD) is one of the best methods for condition monitoring of transformers. In this paper, six types of model transformers, instead of artificial defect models, are designed and manufactured to implant five different defects (scratch on winding insulation, bubble in oil, moisture in insulation paper, very small free metal particle in transformer tank, fixed sharp metal point on transformer tank) on each model, separately. The Continuous Wavelet Transform (CWT) applied to each related measured time-domain PD signal, results in an image representing each single PD signal in time-frequency domain. Then, the Gray Level Covariance Matrix (GLCM) is constructed based on the images from the CWT of PD signals. The texture features are extracted from the constructed GLCM of each PD signal. Also, the principal component analysis are applied to decrease the feature spaces and six first principal components are considered as inputs of the support vector machine for classifying each type of defect models. Results indicate the efficiency of the proposed methods with accurate distinguishing type of defects.

Partial Discharge (PD) is one of the best methods for condition monitoring of transformers. In this paper, six types of model transformers, instead of artificial defect models, are designed and manufactured to implant five different defects (scratch on winding insulation, bubble in oil, moisture in insulation paper, very small free metal particle in transformer tank, fixed sharp metal point on transformer tank) on each model, separately. The Continuous Wavelet Transform (CWT) applied to each related measured time-domain PD signal, results in an image representing each single PD signal in time-frequency domain. Then, the Gray Level Covariance Matrix (GLCM) is constructed based on the images from the CWT of PD signals. The texture features are extracted from the constructed GLCM of each PD signal. Also, the principal component analysis are applied to decrease the feature spaces and six first principal components are considered as inputs of the support vector machine for classifying each type of defect models. Results indicate the efficiency of the proposed methods with accurate distinguishing type of defects.