Research On Bearing Fault Diagnosis Based On Deep Learning And Improved Dispersion Entropy

In industrial production,rolling bearings,as key components of various mechanical equipment,bear different degrees of load generated by the machine during operation,so various faults will inevitably occur.Early fault diagnosis of rolling bearings can effectively reduce the occurrence of downtime and personal accidents.Based on the dispersion entropy method and deep learning theory,this thesis takes the vibration signal of rolling bearings as the research object,and gradually excavates the hidden fault information in rolling bearings under different working conditions from shallow to deep.The fault research methods of characteristic parameters,deep models and cross-domain migration diagnosis are proposed successively to realize the early prediction of bearing faults.(1)Aiming at the problem of non-stationary and large noise interference of rolling bearing signals,based on the nonlinear parameter dispersion entropy theory,an improved multivariate hierarchical multi-scale dispersion entropy method is proposed.The improved multi-scale analysis is combined with the signal hierarchical analysis.Based on the multivariate dispersion entropy theory,various fault features hidden in the rolling bearing signal are extracted.The extracted fault feature data set is reduced by the maximum correlation minimum redundancy method to reduce the dimension and reduce redundancy.Finally,the SVM model is constructed to study the early fault diagnosis of bearings.(2)Aiming at the problem that the fault information under variable load conditions cannot be effectively identified when the fault signal of rolling bearing is extracted,an improved multi-scale analysis and convolutional neural network are proposed to realize the automatic extraction and classification of fault features.The CBAM attention module is used to improve the local feature extraction ability of the model,and the deep hyperparametric convolution is used to improve the diagnostic accuracy while maintaining the parameter quantity.This method uses multi-scale coarse-grained calculation method to down-sample the signal.Based on this,CWT is used to transform the signal into two-dimensional time-frequency diagram,and convolutional neural network is used to classify rolling bearing faults under different working conditions.(3)Aiming at the problem that the training time of the current deep model fault diagnosis process is too long,and it is not suitable for training under non-independent identically distributed samples and small data samples,based on the transfer learning theory,the source domain data is used to pre-train the model parameters,and the model-based transfer strategy is used to solve the problem of the difference between the target domain and the source domain data classification tasks to achieve cross-domain diagnosis.Compared with the traditional fault diagnosis model method the fault diagnosis method based on model-based transfer overcomes the shortcomings of the traditional deep model that requires a large number of data samples for training when performing fault diagnosis.The model has higher training speed and diagnostic accuracy,and is more suitable for actual scenarios.