Because of the robustness of one dimensional time series reconstruction G-P algorithm to extract the fault omen is poor

especially influenced by a sensitive noise in the measured signal.A noise reduction method is proposed based on detrended fluctuat ion analysis（FDA） and kernel principal component analysis（KPCA）

the eigenvalue extraction algorithm based on Mult ifractal spectrum is presented.Through pseudo-phase portrait to determine the weighting factor threshold

optimize and choose parameter and compare with the defects of single G

P

algorithm

and combine with the 3 kinds of rotor system common faults

the stability and accuracy of eigenvalue extraction of this method is analyzed

the results prove that

the diagnosis result is good.