Abstract:The failure of rolling bearing is the foremost cause of the failure and breakdown of rotating machines.As the rolling bearing vibration signal is of nonlinear and nonstationary characteristics,using common time domain or frequency domain approaches cannot easily make an accurate estimation for the rolling element bearing healthy condition.A rolling bearing fault diagnostic approach based on multi-dimensional feature extraction is proposed.Firstly,a multi-dimensional feature extraction algorithm on the basis of entropy,Holder coefficient and improved generalized box-counting dimension theories is proposed for extracting health status feature vectors based on the bearing vibration signals,and secondly a gray relation algorithm is employed for achieving accurate estimation of different fault types and different severities intelligently using the extracted feature vectors.The approach can efficiently and effectively recognize different fault types and different severities in comparison with the existing artificial intelligent methods,and can be suitable for on-line health status monitoring.