Research On Method Of Rolling Bearing Fault Diagnosis Based On Mathematical Morphology And Convolutional Neural Network

Rolling bearings,as a key part of rotating machinery equipment,they affect the operating state of the whole machanical equipment.If a failure occurs,it will cause huge losses.Therefore,it is necessary to carry out fault diagnosis on the running state of rolling bearings.The emphasis is on fault feature extraction and pattern recognition.In order to accurately and quickly extract bearing periodic instantaneous frequency features and realize automatic classification,the fault diagnosis method of rolling bearing based on mathematical morphology and convolutional neural network is studied.The main research contents are as follows:(1)Based on the research of basic morphological theory,an improved single-scale morphological filtering method(ISMF)is proposed.On the basis of studying the properties of basic morphological operators,a new morphological operator—average combined difference filter(ACDIF)is constructed to extract periodic pulse features.Furthermore,the ISMF method is proposed,which takes ACDIF as the morphological operator.At the same time,in order to avoid the influence of the length of the structuring element(SE)on the feature extraction,the fault characteristic frequency ratio(CFR)index is introduced to optimize the length of SE.Through numerical experiments,the ability of ISMF method,other common morphological operators,traditional empirical mode decomposition and wavelet analysis method to extract the periodic pulse features of bearing is studied.(2)In order to remedy the problems of the single-scale morphology method that the bearing fault features cannot be fully extracted due to the single scale and poor noise reduction ability,an adaptive weighted multi-scale morphological filtering method based on improved complete ensemble empirical mode decomposition with adaptive noise(ICEEMDANAWMMF)is proposed.The method uses ICEEMDAN to decompose the original signal to reduce noise,and select components that cover more fault information to reconstruct the signal.ACDIF is used as the morphological operator to analysis the reconstructed signal,then determine the multi-scale optimal range and use the particle swarm optimization algorithm to adaptively generate the weight coefficients of the filtering results of different scales for weighted summation.Numerical experiments and engineering application research show that this method overcomes the shortcomings of ISMF and has better fault feature extraction capabilities than the two traditional multi-scale morphological methods.(3)In order to meet the requirements of fault feature extraction accuracy and real-time performance in actual engineering applications,and to overcome the disadvantages of ICEEMDAN-AWMMF method due to the introduction of noise reduction and optimization algorithms that lead to low computational efficiency,an enhanced scale morphological filtering method(ESMF)is proposed.This method constructs a new combined difference multiply operator(CDMO)to improve the feature extraction ability.On the basis of CDMO filtering,combined with third-order cumlant slice spectrum technology to enhance the fault features and extract the fault feature frequency and its mulplication accurately and quickly.Numerical experiments and engineering applications indicate that this method can enhance the fault features and the time is greatly shortened compared with the ICEEMDAN-AWMMF method.(4)Based on the superiorities of the deep learning method,the ESMF method and convolutional neural network(CNN)are combined to realize automatic classification of bearing faults and meet the needs of intelligent bearing fault diagnosis.This method preprocesses the the complex original vibration signal through ESMF to realize the noise reduction and highlight the fault features.A multi-layer network of CNN is builed to learn the inherent abstract characteristics of the fault signal layer by layer,and realize the mapping from the fault feature domain to the fault category domain,thereby realizing the automation and intelligence of bearing fault diagnosis.It is proved through engineering application that the ESMF method is introduced to preprocess the data,which further improves the fault diagnosis accuracy and robustness of the intelligent classification model.