Classification of Insulation Defects Based on MFCC Coefficients

Transformers are crucial components of power systems, and their insulation status affects the safety and stability of the entire system. To realize the classification of transformer insulation defects based on acoustic signals, this paper presents a method for classifying acoustic signals based on the mel-frequency cepstral coefficient (MFCC) feature extraction technique. First, the MFCC coefficients based on the discrete wavelet transform (DWT) are extracted from the acoustic signals to overcome the weakness of the discrete Fourier transform (DFT). Then, the features are downscaled by using the principal component analysis (PCA) to obtain 1-dimensional static MFCC coefficients and 5-dimensional dynamic ΔMFCC coefficients. Experimental results show that after inputting these features into the support vector machine (SVM) classifier, the classification of the acoustic signals generated by different insulation defects can be realized. The results indicate that the proposed method provides theoretical support for promoting insulation defect classification of transformers based on the acoustic signal detection technology.