Abstract: A fault feature frequency extraction algorithm based on sparse decomposition theory was proposed to solve the problem of chaotic orbit of the rotor directly synthesized． According to the signal characteristics of the ro-tating machinery，the cosine wave model was used to construct a complete dictionary sets，and the matching pursuit algorithm was used to solve the sparse coefficient． The algorithm was used to purify the simulation axis trajectories polluted by white Gaussian noise． The results were almost identical with the ideal ones，and the validity of the algo-rithm is verified． Finally，the proposed algorithm was used to purify the multi-condition axis trajectories under vari-ous of large sliding bearing test-rig，and the corresponding fault types of the rotor were successfully identified． In addition，combined with the theory of dynamic pressure lubrication and the changing rule of spindle position，the fault causes were analyzed．

Key words: sliding bearing, feature extraction, sparse decomposition, axis trajectory, fault diagnosis