Sound anomaly detection of industrial products based on MFCC fusion short-time energy feature extraction

Abstract

Bearing, gear and traditional parts play an important role in the whole mechanical field, and the probability of failure is much higher than that of other mechanical structures, so it is particularly important to carry out state detection and fault diagnosis for such parts. In this paper, a feature extraction method based on Mel Frequency Cepstrum Coefficient (MFCC) fusion of short-time energy features is proposed, and Deep Neural Networks (DNN) is used to identify whether the sound of industrial products at work is abnormal. In this paper, due to the addition of short-term energy information, the information of speech signals can be more accurately obtained, which has better performance than MFCC feature extraction.