Abstract: We used the analysis methods which are based on nonstationary signal analysis methods to study the model and algorithm of the roller bearing fault diagnosis. In the full analysis of failure mechanisms and characteristics of the premise, we focused on the wavelet packet analysis of the vibration signal of rolling bearing fault to extract the effective fault characteristics which could reflect the failure modes. We established the BP neural network classifier based on the fault eigenvectors which we have obtained to achieve recognition and diagnosis of the typical failures of rolling bearings．

Key words: wavelet packet, fault diagnosis, BP neural network, MATLAB