Incipient fault characteristic extraction of rotary machine base on wavelet transform and fuzzy wavelet threshold denoising

Due to the weak energy and nonstationarity, incipient fault characteristic signals are usually submerged by vibration signals of rotary machine and noise. Based on the multi-resolution feature and time-frequency localization feature of Wavelet Transform, a method to extract fault characteristic signals by decomposing them into corresponding time-frequency segmentations is presented. The noise is attenuated, and the characteristic signals are amplified since of the different singularity feature in Wavelet Transform. At the time-frequency segmentations including higher order harmonic frequencies of fault vibration signals, the incipient fault characteristics are extracted efficaciously. The fault signals are denoised further more by a wavelet fuzzy threshold denoising constructed. Higher SNR is gained compared to traditional denoising methods. And the legible time and frequencies fault emerging of characteristic signals are extracted, which can be used to diagnose the position and fault degree combined with the energy of branch reconstruction of fault characteristic signals.