Abstract: The fault feature extraction based on wavelet transform technique is studied in depth，which is applied to fault diagnosis of rotary machine.A new fault feature extraction method based on singularity detection and the modular maximum using continuous wavelet transform analysis is put forward.Research on combining the resolution method and singularity theory can find that wavelet transform is very propitious to depict the whole and local signal.Utilizing the characteristics of wavelet transform resolving and constructing the signal，the information about frequency band pointedly can be chosen and the noise interference can be reduced.At the same time，the typical characteristics of crack faults can be effectively extracted through frequency spectrum analysis of constructing signals.The results show that it is effective to diagnose the non-steady of rotor by wavelet transform.