（ 重庆大学数学与统计学院，重庆 401331； ）
SUN Shanshan， HE Guanghui*
（ School of Mathematics and Statistics, Chongqing University, Chongqing 401331； ）
Fault diagnosis of rolling bearing is very imporant for preventing catastrophic accidents. Due to the the fault vibration signal of rolling bearing is usually non-stationary, and the strong noise interference is contained in the vibration signal at the same time, so effective signal processing techniques are in necessary demands to extract the fault features contained in the collected vibration signals. A fault feature extraction technique based on signal-adapted overcomplete rational dilation discrete wavelet transform is proposed in this paper which allows us to construct a wavelet directly from the statistics of a given signal. And then decompose the input signal to various high frequency band signals by the wavelet bases. Subsequently compute the kurtosis values of the all the high frequency band signals. Then select the optimal signal bands based on maximization of kurtosis value. Finally, the fault features of the optimal signal band is detected through its Hilbert instantaneous frequency spectrum. The experimental results demonstrate the feature extraction technique successfully identifies the incipient fault features.