Research On Fault Prediction Method Based On Spectrum Characteristics

At present,in the field of rotor fault prediction,it is mainly focused on the prediction of the rotor life,but this cannot quantify the development form and failure degree of the rotor fault trend.The rotor spectrum is intuitively and quantitatively reflected the vibration intensity and fault type of the rotor.The quantitative prediction of the rotor's development trend can be realized by the prediction of the rotor spectrum.When the rotor fails,some characteristic frequency amplitudes in the spectrum will increase obviously.In theory,the health state of the rotor can be reflected by monitoring the amplitude changes of these characteristic frequencies,but it needs to know the size of the characteristic frequency beforehand.Prediction of complete spectrum can avoid characteristic frequency selection problem.The full vector spectrum technology can effectively integrate the two channel information of the rotor,avoid information omission,and get a complete signal spectrum.Phase space reconstruction technology can reorganize the related factors of complex systems and describe the whole system with finite data.The extreme learning machine algorithm has simple network structure and it has fast computing speed and high learning efficiency.Therefore,combining the advantages of the two,the spectrum structure prediction model based on the extreme learning machine and phase space reconstruction is proposed,and it can realize the prediction of the complete spectrum structure.The main contents of this paper are as follows:(1)Propose a full vector information fusion method based on multidimensional empirical mode decomposition(MEMD)which is called FV-MEMD.First,the collected homologous dual channel signals are decomposed into a series of IMFs using the MEMD algorithm;then the IMFs,which contains the main information of the rotor,are reconstructed according to the correlation coefficient criterion to achieve the purpose of reducing the noise;at last,calculate the spectrum features of the reconstructed signal by full vector spectrum.The effectiveness of the method is verified by simulation analysis and case analysis.(2)A multivariable reconstruction method based on spectrum features is studied.Based on the theory of phase space reconstruction,a multivariable phase space reconstruction model is established,and the predictability of the reconfiguration model of the extreme learning machine network is discussed,and the input and output expressions of the prediction model are established.The simulation results show that the proposed method is effective.(3)The frequency spectrum prediction model of the extreme learning machine based on phase space reconstruction is proposed,and the evaluation index of the prediction results is given.First,the training and test samples are established by the multivariable and multi output reconfiguration model with the spectrum data;the samples are then trained and predicted in the extreme learning machine;finally,the whole life monitoring data of rolling bearing is used to verify.The results show that the method can accurately predict the spectrum structure of bearing.