Application of singular value decomposition and short time Fourier transform in sound information separation

In order to deal with problems in low signal-to-noise ratio signal analysis and feature extraction of cycle-nonstationary signal, a method based on singular value decomposition and short time Fourier transforms (STFT) is adopted for the signals collected from the rolling mill with sound sensors. In the practical data processing, the method can reduce the noise in the signals so that the useful information is separated out and the diagnosis quality can be improved. The reconstructed signal compared with original signal shows better performance for the feature extraction by using STFT.