The improved Huber M method which fuse to two robust processings of median and Huber M method was proposed to solve the robust processing of data. The similarity of the median and average value was given to judge whether the variation data exist

the variation rate were determined by trend of the variation rate. Within the variation rate range of 0%-10%

the continuity and credibility of the rolling bearing vibration data increase with the increasing of the variation rate. The dynamic characteristics of rolling bearings were analyzed by using chaos theory. Time delay

embedding dimension and the maximum Lyapunov index of vibration data are the same on the same batch of rolling bearing. And the maximum Lyapunov index which is bigger than 0

show that the dynamic characteristic on rolling bearing vibration belongs to chaos characteristics

moreover the maximum forecast cycle are calculated

namely 667 unit. The nonlinear and monotonicity of the median of physical space and estimated correlation dimension of phase space on rolling bearing vibration time series is the inner running mechanism on rolling bearing vibration

which provide reliable basis for further analysis.