Research On Fault Diagnosis Of Rolling Bearing Base On SGMD And Neural Network

With the rapid development of industrial informatization,mechanical equipment is developing towards the direction of intelligence,networking and complexity,and its integration and precision are constantly improving.As an important supporting part of rotating machinery,rolling bearing plays a vital role in the stable operation of equipment.Generally,the rolling bearing is closely connected with the shaft,bearing seat and other parts.When the rolling bearing fails,their vibration signals are often coupled by multiple vibration signals and noises.Therefore,it is often difficult to extract fault features from vibration signals of rolling bearings,which leads to errors in fault feature extraction and fault classification.In view of the above problems,this thesis studies the vibration signal processing method and fault identification method of rolling bearings,and puts forward a fault feature extraction method—Symplectic Geometry Mode Decomposition Based on Similarity and Kurtosis(SGMD-SK)and a fault identification method based on one-dimensional residual network of convolution attention module.The specific work of this thesis is as follows:(1)The signal decomposition method is studied,especially SGMD method.Comparing SGMD with EMD,LMD and VMD,the superiority of Symplectic Geometry Mode Decomposition method in anti-modal aliasing and anti-noise is verified by simulation and experimental analysis.(2)The kurtosis value of the signal is large,but SGMD only relies on the similarity criterion for component recombination,without taking kurtosis index into account.In order to make the reconstructed components contain more impact features,this thesis proposes SGMD-SK,which uses the comprehensive index of similarity-kurtosis to improve SGMD,and the effectiveness of this method is verified by experiments.(3)The vibration signal of rolling bearing contains a lot of noise,and the single-channel vibration signal is difficult to fully reflect the fault information.To solve the above problems,this thesis further proposes a fault feature extraction method based on full vector SGMD-SK and FastICA,which can reduce the noise in the signal,enhance the impact feature and fully reflect the fault features of rolling bearings.(4)The fault identification method based on feature vector has some problems,such as complex feature extraction of rolling bearing fault,difficult description of complex relationship between signals,low accuracy of fault identification and complicated process.In order to solve the above problems,this thesis proposes a one-dimensional residual network model ResNet10 and realizes end-to-end fault identification.In order to further improve the accuracy,a one-dimensional ResNet10 network model based on convolution module attention mechanism is proposed.The experimental results show that the fault identification accuracy of the rolling bearing of this model is further improved to 99.86%.The fault identification of rolling bearings under different loads can still achieve high accuracy.