Research On Wavelet Analysis And Machine Learning Based Rolling Bearing Fault Diagnosis

With the continuous advancement of national industrialization,the degree of mechanical equipment automation in factories is getting higher and higher,and the research on fault diagnosis technology of mechanical equipment is becoming increasingly important.As one of the most widely used and easily damaged parts in rotating machinery,the complexity and non-stability of vibration signal promote the fault diagnosis technology of rolling bearing to become more and more intelligent.This paper mainly studies two aspects of fault feature extraction and pattern recognition in rolling bearing fault diagnosis technology:In the aspect of fault feature extraction,this paper explores the theory of wavelet analysis firstly.Then,wavelet analysis is used to denoise the vibration signal extracted from the bearing fault diagnosis test-bed,and wavelet packet transform is used to extract the energy features of the noise reduced signal.The energy features are used as the fault feature vector in the subsequent pattern recognition.In the aspect of fault pattern recognition,this paper studies the application of BP neural network,Support Vector Machine(SVM)and Stacked Sparse Autoencoder network(SSAE)in machine learning.Firstly,the working principle of BP neural network is studied,and the network structure and network parameters are determined according to the fault data.The data in the fault sample set are randomly selected to train and test the network,and the experiment has obtained the good results.Secondly,the paper studied the classification performance of SVM,explored the influence of different kernel functions and kernel parameters on the classification effect,and then applied it to the fault diagnosis of bearing with remarkable results.In view of the fact that the above two methods are both supervised learning methods,which need to label the input data and fault types at the same time to achieve classification.Therefore,a new intelligent unsupervised fault diagnosis method based on stack sparse autoencoder is proposed in the end of this paper.In this method,the energy features extracted by wavelet packet is used as the input of the network.The paper also analyzes the influence of the number of nodes in different hidden layers on classification performance of network.After determining the network structure and parameters,the classification effect of the model on bearing normal state,inner ring pitting state and outer ring pitting state is investigated.By comparing the diagnosis results of BP neural network and SVM,the correctness and validity of the method are verified.