A Method Of Rotor System Fault Diagnosis Based On Adaptive And Sparsest Time-Frequency Analysis

The key of rotor system diagnosis is to extract fault features effectively.The feature extraction method based on time-frequency analysis can reveal the information embedded in the signal’s time-frequency distribution,which has been applied widely in the fault extraction field.Based on this,this paper introduces a new time-frequency method-Adaptive and Sparsest Time-Frequency Analysis(ASTFA).on the basis of theory research and improvement in ASTFA,combined with ASTFA,singular value decomposition,entropy theory and neural network,the new application in rotor system fault diagnosis is proposed.The main contents of this paper are as follows:1.Two simulation examples are analyzed,and the results show the fact that ASTFA works better in restraining boundary effect than empirical mode decomposition,and in noise resistance better than ensemble empirical mode decomposition.2.The initial phase function of θ0 and bandwidth parameter of λmax are discussed in details to demonstrate their impact on ASTFA.Aimed at the problem that improper,Amax may cause mode mixing,the paper puts forward complete adaptive and sparsest time-frequency analysis(CASTFA).CASTFA can determines the λmax adaptively using EEMD method to make ASTFA’s adaptability more complete.Simulation example and the practical case of rotor rubbing fault diagnosis both manifest that CASTFA performs better than ASTFA in restraining mode mixing.3.Two kinds of feature extraction method are proposed.One is based on CASTFA and singular value entropy,and the other is based on CASTFA and time-frequency entropy.The application examples of both methods in rotor fault diagnosis prove their efficiency.4.A systematic method of rotor fault diagnosis is proposed,which combines CASTFA with BP neural network.Firstly,CASTFA is adapted to decompose the vibration signal into several components;then AR mode is used to extract features of those components;finally,AR mode parameters are regarded as input of BP neural network,and fault classes as output of BP neural network.A practical example proves that the method works well in rotor system fault diagnosis.