The vibration signal of rolling bearing is a nonlinear and unstable signal. Therefore it is very challenging to carry out feature extraction accurately from the complicated data of non-periodic rolling bearing. This article hereby proposes a method of feature extraction based on local spectrum bearing data. This method combined the segmentation point obtained from pretreatment and the spectrum analysis

built localized feature of the data

determined the definition of the local frequency and the construction method of time-frequency domain

and implemented the feature extraction. Experiments show that this method overcame the limitation that Hilbert transform is only suitable to describe the narrowband signals. It also made up for the defects of Fourier global frequency which is only valuable to the infinite wave period signals. As a new method of feature extraction from the time domain data of the nonlinear and unstable rolling bearing

it reduces the false frequency and is compatible with the analysis of both time domain and frequency domain. It has very high practical value in the fault diagnosis of rolling bearings.