Application of Ensemble Learning Method Based on Wavelet Packet Decomposition in Bearing Fault Diagnosis

This paper proposed a diagnosis method based on wavelet packet decomposition and improved ensemble learning (IEL) for the rolling bearing. Firstly, the vibration signal of rolling bearing is decomposed by three layers of wavelet packet. Then four low-frequency components and four high-frequency components are extracted. Secondly, the data set is constructed by adding the square sum of the extracted components. Next, IEL algorithm determines a unified F
1
parameter threshold for each diagnosis model, and then obtains the F
1
parameter of each diagnosis model for single fault through experiments. Finally, the F
1
parameter is compared with the threshold value, and the weight coefficient of IEL is constructed to form the final diagnosis model. The proposed method is superior to the model constructed by each diagnosis model separately and the general ensemble learning method.