Abstract:

In order to monitor the healthy operation state of the roller, the normal audio signal and the abnormal audio signal of the roller were extracted through the field test. Since the extracted audio signal contains a lot of noise, an improved wavelet threshold denoising method was proposed in this paper. This method effectively filtered out the noise in the audio signal and laid a foundation for the later feature extraction of the signal. In order to further study the characteristic difference between normal audio signal and abnormal audio signal, the MFCC feature extraction method was used to obtain an MFCC feature characterization map which could obviously show the difference between normal state and abnormal state of the roller. The results show that the time domain diagram and spectrum diagram of the abnormal audio signal fluctuate more violently than the normal audio signal; the MFCC feature characterization map of the normal audio signal of the roller is higher than the initial amplitude of the abnormal audio signal, and the amplitude decreases more slowly.

Key words: roller fault, abnormal audio, wavelet threshold denoising, filter, Mel-frequency cepstrum coefficient (MFCC)