Bearing Fault Diagnosis Based On Multi Feature Fusion And Deep Belief Network

Mechanical fault diagnosis is of great significance to guarantee the safe and stable operation of mechanical equipment and one of the important branch of the mechanical fault diagnosis is bearing fault diagnosis.In this paper,a new fault diagnosis method based on multi feature fusion and deep belief network is proposed to deal with the complex and nonlinear characteristics of vibration signals.Firstly,the bearing fault feature extraction method based on dual tree complex wavelet(DTCWT)is studied.It is pointed out that DTCWT has the excellent characteristics of shift invariance and suppression of frequency aliasing,and can effectively decompose the mechanical vibration signal.The fault diagnosis experiment of bearings with different locations is carried out,and the bearing fault diagnosis is realized by fuzzy c-means(FCM)clustering.Which verifies the validity of the method of bearing fault feature extraction.Secondly,the bearing fault feature extraction method based on multi-masking empirical mode decomposition(MMEMD)is studied.An improved algorithm MMEMD is proposed for the EMD mode mixing.The simulation results show that MMEMD can effectively solve the mode mixing problem of EMD,which is helpful for bearing vibration signal decomposition.Combined with support vector machine(SVM),the fault diagnosis experiments are carried out for 10 kinds of bearing states,including normal state,different position and different degrees of damage.Which verifies the validity of the method of bearing fault feature extraction.Thirdly,the bearing fault feature extraction method based on VMD is researched.VMD uses the variational framework to solve the EMD modal aliasing problem,and can realize the signal filtering,which is helpful to the bearing fault feature extraction.the fault diagnosis experiments are carried out for 10 kinds of bearing states,including normal state,different position and different degrees of damage.Which verifies the validity of the bearing fault feature extraction method.Fourthly,for the problem of mechanical fault recognition,a small sample classifier based on deep belief network(DBN)is designed.Through the classification of Iris data sets,it is proved that the classification accuracy and stability are better than SVM for small sample classification.Combined the VMD sample entropy,bearing fault diagnosis experiment are carried out for 10 kinds of bearing states,including normal state,different position and different degrees of damage.Which verifies its effectiveness in identifying bearing fault mode.Finally,the multi features and DBN are combined to achieve the bearing fault diagnosis.First of all,the bearing vibration signals are analyzed by FCM clustering to determine whether there are faults and the labels of a few samples.Then the DTCWT sample entropy,MMEMD energy entropy and VMD sample entropy are extracted.And fusion characteristics of bearing fault is obtained by principal component analysis(PCA).Finally,the DBN small sample classifier is used to classify the fault,and the fault type is determined by the time-frequency analysis method.The fault diagnosis experiments are carried out for 10 kinds of bearing states,including normal state,different position and different degrees of damage.Which verifies its effectiveness for bearing fault diagnosis.The method is used for fault diagnosis of wind turbine bearing,and its practical value is verified.