Vibration Signal Feature Extraction Based On Dual-Tree Complex Wavelet Transform

With the development of rotating machineries, they have being used widely in modern industry. It is an important matter for fault diagnosis to maintain and detect the key parts of the rotating machineries. Rolling bearing is also a most important segment in rotating machineries, but is vulnerable for frequently working. Therefore, the bearing vibration signal processing is valuable to the fault diagnosis .In the fault diagnosis of bearing vibration signal, feature extraction plays a significant role. Efficient extraction can do a favor to exactly analysis the reasons of the fault and know about the degree of fault defection. In this paper, the feature frequency extraction is proposed by a new technique based on the dual-tree complex wavelet transform (DTCWT). A new approach is also proposed to analyze the results of the decomposition of signals with DTCWT, which obtain the fault state trends.The related basic theories between Fourier transform, wavelet transform and DTCWT are introduced first. Then the properties of DTCWT shift invariance and aliasing resistance show that advantage of DTCWT comparing to DWT. According to the DTCWT filter banks designed and efficient values chosen, there are comparative feature extraction results between the DTCWT and DWT in one dimensional simulation signals. That proves the outstanding practicality for DTCWT in one dimension signal processing.At last, the experimental vibration data are used to do the fault feature frequency extraction experiment by DTCWT. There are three group experiment tests containing inner race fault, outer race fault and ball slide fault. The results show that DTCWT do a good job to extract the feature signal in low frequency. We also get tends of the three fault types according to the DTCWT analysis signals experimental data. This illustrates that DTCWT is a good tool for fault diagnosis in vibration signals analysis and provides a further foundation to fault identifying later.