Fault Diagnosis Methods Of Rolling Bearing Based On Isomap

Rolling bearing is one of the most widely used part in rotating machinery,and its health directly affects the normal operation of the whole mechanical system.Therefore,fault diagnosis and state detection of rolling bearings have important theoretical and practical significance for the healthy operation of mechanical equipment and the avoidance of major accidents in the production process.Since most of rolling bearing vibration signals are non-stationary and nonlinear,it is necessary to extract fault features from multiple aspects to characterize its operating state.Thus,the fault features set has high dimensionality,non-linearity,redundancy and other characteristics,which leads to the problem of "curse of dimensionality".As a kind of nonlinear dimensionality reduction method,manifold learning can effectively find low-dimensional manifold structures embedded in high-dimensional data space,and realize dimensionality reduction or data visualization.Hence,it is suitable for dimensionality reduction of high-dimensional fault features of rolling bearings.On the basis of discussing the status of of signal de-noising,feature extraction,dimensionality reduction and pattern recognition for the mechanical equipment,this thesis focuses on the study of the isometric mapping(Isomap)method and its improved methods for fault features dimensionality reduction.At the same time,some key technical problems involved in the process of fault features extraction and fault diagnosis of rolling bearings have been deeply studied,and some research results have been obtained.The main research and innovation results of this dissertation are as follows:(1)Aiming at solving the problem of neighborhood construction and landmarks selection in L-Isomap,a novel manifold learning method based on adaptive self-organizing incremental neural network landmark Isomap(ASL-Isomap)was proposed.Via combining preserving the consistency of global manifold information,adaptive neighborhood construction and SOINN landmarks selection,ASL-Isomap could gain a perfect approximation of low-dimensional intrinsic geometric structure within the high-dimensional observation data,and the experimental results showed that it has better dimensionality reduction than the existing dimensionality reduction method.In view of the unknown label information of rolling bearing faults,a rolling bearing fault diagnosis method based on hybrid domain features set,ASL-Isomap and kernel extreme learning machine(KELM)was proposed.This method integrated the advantages of hybrid domain fault feature set for fault information mining,ASL-Isomap for automatic dimensionality reduction and KELM for pattern recognition,and could effectively and accurately diagnose various fault types of rolling bearings.The fault diagnosis experiment of 4210 cylindrical roller bearings verifies the effectiveness of the proposed method.(2)Based on the SOINN L-Isomap(SL-Isomap)algorithm and the supervised learning,a novel supervised SOINN L-Isomap(SSL-Isomap)manifold learning method was proposed.On the one hand,the SOINN method was used to automatically generate a series of landmarks that could represent the original data topology,which solved the problem of L-Isomap landmarks selecting;on the other hand,sample labels were used to construct supervised neighborhood graphs.Thus,it had better further features extraction and fault classification accuracy.Aiming at the known label information of rolling bearing faults,a fault diagnosis model of rolling bearing based on multi-scale permutation entropy(MPE),SSL-Isomap and grey wolf optimizer support vector machine(GWO-SVM)was proposed.The fault diagnosis analysis of 4210 cylindrical roller bearing showed that the proposed fault diagnosis method integrates MPE in feature extraction,SSL-Isomap in dimensionality reduction and the GWO-SVM in the pattern recognition,and can accurately determine the different fault types.(3)Combining SL-Isomap with the the semi-supervised learning algorithm,a novel semi-supervised SOINN L-Isomap(SSDSL-Isomap)manifold learning method was proposed.On the one hand,semi-supervised doubles neighborhood graph was constructed by using labeled samples and unlabeled samples,which enhanced the data classification ability;on the other hand,the SOINN method was used to the landmarks selection,which overcame the defect that the traditional stochastic landmarks selection method led to unstable dimensionality.And then,the proposed method was applied to the dimensionality reduction process of high-dimensional fault feature set.In view of the known label information of part of rolling bearing faults,a fault diagnosis method of rolling bearings based on improved composite multiscale sample entropy based on variational mode decomposition(VMD-ICMSE),SSDSL-Isomap and particle swarm optimization extreme learning machine(PSO-ELM)was developed.The experimental results of 1210self-aligning ball bearings fault diagnosis showed that the proposed method integrates VMD-ICMSE in feature extraction,SSDSL-Isomap in dimensionality reduction,and PSO-ELM in pattern recognition,and can effectively and accurately diagnose the fault types.