Mechanical Fault Diagnosis of GIS Based on MFCCs of Sound Signals
Chen Honggang, Xu Mingyue, Fu Chenzhao, Song Renjie, Zhe Li
Abstract
Mechanical fault diagnosis of Gas Insulated Switchgear (GIS) is a popular research topic in recent years. Inspired by the fact that experienced inspectors can detect the failure of GIS only by sound, this paper proposes a novel framework for mechanical fault diagnosis based on sound classification. We extract the Mel Frequency Cepstral Coefficient (MFCC) features from the input sound signal and make an improvement to adapt to the underlying features of the GIS’s sounds. Using machine learning, we classify the extracted features with a support vector machine (SVM) to obtain a model of mechanical fault diagnosis of GIS based on sound signals. We outperform the traditional MFCC features by a significant margin, especially when the input sound signal is disturbed by noises. The F1-score can be increased by 2.47% to 21.59% with the change of the signal-to-noise (SNR) ratio.