Research On Fault Diagnosis Of Motor Bearing Based On Feature Fusion

China is a big industrial manufacturing country.With the rapid development of various industries,motor as a power equipment,plays an increasingly important part,especially in the normal operation of the motor in the production line.Among all the fault types of motor,bearing fault accounts for about 40%,which is easy to cause accidents.Therefore,this paper studies the fault diagnosis algorithm of motor bearing,the main work is as follows:(1)Firstly,the vibration signals of motor bearing under normal condition and in case of fault(inner ring fault,outer ring fault,ball fault)are extracted.Then,this paper use complementary set empirical mode decomposition,wavelet packet decomposition and timedomain parameter extraction to extract the characteristics of vibration signals,combining the extracted results with the information concepts such as permutation entropy and energy entropy,Finally,three kinds of different characteristic parameters(complementary set EMD permutation entropy,wavelet packet energy entropy and time domain parameters)are obtained(2)In this paper,feature selection and feature fusion are carried out for the extracted feature quantity.Firstly,feature selection is needed after feature quantity is extracted from vibration signal,so genetic algorithm is used to select the best feature among all kinds of features.Then,all kinds of optimal features are fused by generalized discriminant analysis to get low dimensional fusion features.Support vector machine is used to identify single type feature,combined feature and fusion feature respectively,the experimental results show that the diagnosis rate of all kinds of faults can reach 100%.(3)Using convolution neural network as classifier,the fault diagnosis of motor bearing is studied from the perspective of deep learning,Firstly,one-dimensional vibration signal is transformed into two-dimensional wavelet time-frequency characteristic map by wavelet transform,Then the wavelet time-frequency characteristic map is input to convolutional neural network for fault diagnosis,Finally,the diagnosis rate of all kinds of faults is 100%,which also has a good fault diagnosis effect.