Abstract:Rotating mechanical axisorbit contains the rich information of its running state, which is the important basis for identification of rotor running state and fault sign.Invariant moment and Fourier descriptorare proposed to beutilized to extract the axisorbit image features.The DS evidentialtheory was adopted to orbit characteristic parameters into the fusion and diagnosisidentification,andthen compared with the traditional BP neural network identification method, the result shown that the DS evidential theory improved the identification accuracy.The proposed method was applied to magnetic bearing fault diagnosis, the measured vibration signal was usedto verify the practicability of this method,and the final results showed that the identification result was inconsistent with orbit shape.It isproved that the proposed method can not only extract axisorbitimage features, but also effectively carry outtheaxis orbitidentification to improve the precision of the magnetic bearing fault diagnosis.